diff --git a/.github/scripts/install-ubuntu-apt-packages.sh b/.github/scripts/install-ubuntu-apt-packages.sh new file mode 100644 index 00000000000..b5790f3f90d --- /dev/null +++ b/.github/scripts/install-ubuntu-apt-packages.sh @@ -0,0 +1,28 @@ +#!/usr/bin/env bash +set -euo pipefail + +if [ "$#" -eq 0 ]; then + echo 'Usage: install-ubuntu-apt-packages.sh ...' >&2 + exit 2 +fi + +bash "$(dirname "${BASH_SOURCE[0]}")/prepare-ubuntu-apt.sh" + +# Bound both routes while allowing large fallback downloads. +# Run deadlines as root so sudo cannot leave apt in a separate session. +if sudo timeout --kill-after=10s 45 apt-get update -o Acquire::Languages=none && sudo timeout --kill-after=10s 120 apt-get install -y "$@"; then + exit 0 +fi + +echo 'Ubuntu package setup failed; trying the same packages through the MIT mirror.' >&2 +for host in archive.ubuntu.com ports.ubuntu.com mirrors.mit.edu; do + timeout 5 getent ahosts "$host" || true +done + +bash "$(dirname "${BASH_SOURCE[0]}")/use-ubuntu-apt-mirror.sh" +bash "$(dirname "${BASH_SOURCE[0]}")/recover-ubuntu-dpkg.sh" || true +# Unrelated repositories may fail refresh; apt still verifies the requested packages. +if ! sudo timeout --kill-after=10s 120 apt-get update -o Acquire::Languages=none; then + echo 'Index refresh incomplete; attempting the requested packages with available authenticated indexes.' >&2 +fi +sudo timeout --kill-after=10s 600 apt-get install -y "$@" diff --git a/.github/scripts/recover-ubuntu-dpkg.sh b/.github/scripts/recover-ubuntu-dpkg.sh new file mode 100644 index 00000000000..e57a094aa76 --- /dev/null +++ b/.github/scripts/recover-ubuntu-dpkg.sh @@ -0,0 +1,10 @@ +#!/usr/bin/env bash +set -euo pipefail + +# A timed-out apt frontend can leave dpkg running in a separate session. +sudo timeout --kill-after=10s 30 bash -euo pipefail -c ' + while lslocks --noheadings --raw --notruncate --output PATH | grep -E "^/var/lib/dpkg/lock(-frontend)?$" >/dev/null; do + sleep 1 + done + dpkg --configure -a +' diff --git a/.github/scripts/use-ubuntu-apt-mirror.sh b/.github/scripts/use-ubuntu-apt-mirror.sh new file mode 100644 index 00000000000..b3617d59e67 --- /dev/null +++ b/.github/scripts/use-ubuntu-apt-mirror.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash +set -euo pipefail + +# Change repository locations while retaining suites, architectures and signing keys. +for source in /etc/apt/apt-mirrors.txt /etc/apt/sources.list /etc/apt/sources.list.d/*.list /etc/apt/sources.list.d/*.sources; do + if [ -f "$source" ]; then + sudo sed -i -E \ + -e 's#https?://(azure\.archive|archive|security)\.ubuntu\.com/ubuntu#https://mirrors.mit.edu/ubuntu#g' \ + -e 's#https?://ports\.ubuntu\.com/ubuntu-ports#https://mirrors.mit.edu/ubuntu-ports#g' \ + "$source" + fi +done diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml index 9339215a217..ca0f42a63c9 100644 --- a/.github/workflows/e2e.yml +++ b/.github/workflows/e2e.yml @@ -123,6 +123,10 @@ jobs: exit 0 fi read -r -a packages <<< "$ORCA_E2E_APT_PACKAGES" + if [ -f .github/scripts/install-ubuntu-apt-packages.sh ]; then + bash .github/scripts/install-ubuntu-apt-packages.sh "${packages[@]}" + exit 0 + fi if [ -f .github/scripts/prepare-ubuntu-apt.sh ]; then bash .github/scripts/prepare-ubuntu-apt.sh else @@ -529,7 +533,12 @@ jobs: ref: ${{ inputs.ref || github.ref }} persist-credentials: false - name: Install headless tools - run: sudo apt-get update && sudo apt-get install -y build-essential ripgrep xvfb openbox x11-utils + run: &install_headless_tools | + if [ -f .github/scripts/install-ubuntu-apt-packages.sh ]; then + bash .github/scripts/install-ubuntu-apt-packages.sh build-essential ripgrep xvfb openbox x11-utils + else + sudo apt-get update && sudo apt-get install -y build-essential ripgrep xvfb openbox x11-utils + fi - uses: ./.github/actions/install-node-dependencies with: native-runtime: electron @@ -636,7 +645,7 @@ jobs: - name: Fetch the 1.4.224 relay sources run: git fetch --no-tags --depth=1 origin +refs/tags/v1.4.224:refs/tags/v1.4.224 - name: Install headless tools - run: sudo apt-get update && sudo apt-get install -y build-essential ripgrep xvfb openbox x11-utils + run: *install_headless_tools - uses: ./.github/actions/install-node-dependencies with: native-runtime: electron diff --git a/.github/workflows/golden-e2e-experiment.yml b/.github/workflows/golden-e2e-experiment.yml index da3e9e55ef8..d059bd2f42f 100644 --- a/.github/workflows/golden-e2e-experiment.yml +++ b/.github/workflows/golden-e2e-experiment.yml @@ -70,7 +70,6 @@ jobs: - name: Run golden E2E tests on Linux if: runner.os == 'Linux' run: | - xvfb-run --auto-servernum env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run test:e2e tests/e2e/golden-core-flows.spec.ts xvfb-run --auto-servernum env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run --if-present test:e2e:workspace-session-golden if [ -f tests/e2e/golden-fresh-profile-terminal.spec.ts ]; then xvfb-run --auto-servernum env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run test:e2e tests/e2e/golden-fresh-profile-terminal.spec.ts tests/e2e/golden-shell-command.spec.ts @@ -84,7 +83,6 @@ jobs: - name: Run golden E2E tests on macOS if: runner.os == 'macOS' run: | - env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run test:e2e tests/e2e/golden-core-flows.spec.ts env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run --if-present test:e2e:workspace-session-golden if [ -f tests/e2e/golden-fresh-profile-terminal.spec.ts ]; then env SKIP_BUILD=1 ORCA_E2E_FORWARD_APP_LOGS=1 pnpm run test:e2e tests/e2e/golden-fresh-profile-terminal.spec.ts tests/e2e/golden-shell-command.spec.ts diff --git a/.github/workflows/pr.yml b/.github/workflows/pr.yml index bd71e39e97a..ef97a54bf90 100644 --- a/.github/workflows/pr.yml +++ b/.github/workflows/pr.yml @@ -790,15 +790,7 @@ jobs: id: shells background: true run: | - # Per-connection limits complement the whole-command apt deadlines below. - sudo tee /etc/apt/apt.conf.d/99-orca-shell-contracts >/dev/null <<'APTCONF' - Acquire::http::Timeout "15"; - Acquire::https::Timeout "15"; - Acquire::Retries "1"; - APTCONF - # Unrelated runner repositories may fail refresh; installing zsh must still succeed. - timeout 120 sudo apt-get update || true - timeout 300 sudo apt-get install -y zsh + bash .github/scripts/install-ubuntu-apt-packages.sh zsh # A fixed 4.x release meets the contract without tracking package-repository updates. case "$(uname -m)" in aarch64) arch=aarch64 sha=57349f0111728c64b703a84e611aa97ff4d1fa8466385d64e67bd5bc3ffa49fc ;; @@ -936,7 +928,18 @@ jobs: background: true run: | bash .github/scripts/prepare-ubuntu-apt.sh - pnpm exec playwright-core install --with-deps webkit + if [ ! -f .github/scripts/use-ubuntu-apt-mirror.sh ]; then + pnpm exec playwright-core install --with-deps webkit + exit 0 + fi + echo 'Installing WebKit dependencies through the primary route.' + if ! sudo env "PATH=$PATH" timeout --kill-after=10s 240 ./node_modules/.bin/playwright-core install-deps webkit; then + echo 'Dependency installation did not complete; retrying with the Ubuntu mirror fallback.' + bash .github/scripts/use-ubuntu-apt-mirror.sh + bash .github/scripts/recover-ubuntu-dpkg.sh || true + sudo env "PATH=$PATH" timeout --kill-after=10s 600 ./node_modules/.bin/playwright-core install-deps webkit + fi + timeout --kill-after=10s 180 pnpm exec playwright-core install webkit # The entry lives in mobile/ so one React resolves; without this every RN import is nothing. - uses: ./.github/actions/install-mobile-dependencies diff --git a/.github/workflows/ssh-windows-hosts.yml b/.github/workflows/ssh-windows-hosts.yml index 4185ad26af4..9f7a2496700 100644 --- a/.github/workflows/ssh-windows-hosts.yml +++ b/.github/workflows/ssh-windows-hosts.yml @@ -62,7 +62,6 @@ on: - 'src/shared/shell/launch/shell-argument-quoting.ts' - '!src/**/*.test.ts' - 'src/main/managed-server/host-state/orcad-windows-host-lane.test.ts' - - 'tests/e2e/ssh-orcad-auto-convert.spec.ts' - 'tests/e2e/helpers/orcad-convert-host.ts' - 'tests/e2e/helpers/orcad-convert-flow.ts' - 'tests/e2e/helpers/orcad-upgrade-profile.ts' @@ -78,7 +77,7 @@ on: workflow_dispatch: inputs: cells: - description: Comma-separated cell ids from src/main/ssh/hostile-host-test-support/ssh-windows-host-cells.ts; empty runs the default set (orcad-cli-* cells run only when named). + description: Comma-separated cells orcad-cmd, orcad-powershell, or orcad-cli-managed; empty runs the two shell cells. required: false default: '' preparation_mode: @@ -144,7 +143,7 @@ jobs: with: persist-credentials: false # Keep complete server/test trees; missing future imports fail the unchanged builds. - # All of src: the orcad-convert cell builds the full e2e app and the bundled CLI. + # All of src: the managed CLI cell builds the full e2e app and bundled CLI. # mobile: the template ships the mobile web page, which builds from mobile/. # cloud: config/tsconfig.vitest-transform.json references the cloud projects, so vitest needs them. sparse-checkout: | @@ -242,10 +241,8 @@ jobs: New-Item -ItemType Directory -Force -Path $receipts | Out-Null $cells=@($env:CELLS -split ',' | ForEach-Object {$_.Trim()} | Where-Object {$_}) if(-not $cells.Count){$cells=@('orcad-cmd','orcad-powershell')} - # One Windows host runs the app-level conversion; it is last because it switches native modules. - if(-not $env:CELLS -and '${{ matrix.arch }}' -eq 'x64' -and '${{ matrix.server }}' -eq 'inbox'){$cells+='orcad-convert'} # App cells reach the managed server through an SSH local forward, which provisioning grants to the last accounts only. - $appCellIds=@('orcad-convert','orcad-cli-managed','orcad-cli-convert') + $appCellIds=@('orcad-cli-managed') $cells=@($cells | Where-Object {$appCellIds -notcontains $_})+@($cells | Where-Object {$appCellIds -contains $_}) $forwarding=@($cells | Where-Object {$appCellIds -contains $_}).Count $archive='' diff --git a/config/ci/windows-ssh-provider/invoke-windows-host-cells.ps1 b/config/ci/windows-ssh-provider/invoke-windows-host-cells.ps1 index 0483fc0d02a..d4d1c9c4f26 100644 --- a/config/ci/windows-ssh-provider/invoke-windows-host-cells.ps1 +++ b/config/ci/windows-ssh-provider/invoke-windows-host-cells.ps1 @@ -4,13 +4,13 @@ param( [Parameter(Mandatory=$true)][hashtable]$Context, [Parameter(Mandatory=$true)][ValidateSet('win32-arm64','win32-x64')][string]$Target, [Parameter(Mandatory=$true)][string]$ReceiptRoot, - [ValidateSet('orcad-cmd','orcad-powershell','orcad-convert','orcad-cli-managed','orcad-cli-convert')][string[]]$Cells=@('orcad-cmd','orcad-powershell') + [ValidateSet('orcad-cmd','orcad-powershell','orcad-cli-managed')][string[]]$Cells=@('orcad-cmd','orcad-powershell') ) $ErrorActionPreference='Stop' if($env:GITHUB_ACTIONS -ne 'true' -or $env:ORCA_ISOLATED_SSH_CI -ne '1'){throw 'Disposable CI only'} -$shells=@{'orcad-cmd'='cmd';'orcad-powershell'='powershell';'orcad-convert'='cmd';'orcad-cli-managed'='cmd';'orcad-cli-convert'='cmd'} +$shells=@{'orcad-cmd'='cmd';'orcad-powershell'='powershell';'orcad-cli-managed'='cmd'} # App-level cells drive the e2e build (and the bundled CLI) against the host; each greps one tagged test. -$appCells=@{'orcad-convert'=@('tests/e2e/ssh-orcad-auto-convert.spec.ts','');'orcad-cli-managed'=@('tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts','@orcad-cli-managed');'orcad-cli-convert'=@('tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts','@orcad-cli-convert')} +$appCells=@{'orcad-cli-managed'=@('tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts','@orcad-cli-managed')} $electronBuilt=$false if($Context.accounts.Count -lt $Cells.Count){throw 'Each cell needs its own private account'} $seenApp=$false diff --git a/config/flaky-test-pattern-baseline.txt b/config/flaky-test-pattern-baseline.txt index 2e5c54296a7..715f80fa19a 100644 --- a/config/flaky-test-pattern-baseline.txt +++ b/config/flaky-test-pattern-baseline.txt @@ -92,7 +92,6 @@ 1 test-no-polling-loop:tests/e2e/fixtures/daemon-generation-reconnect-client.ts 2 test-no-polling-loop:tests/e2e/floating-workspace-reopen-webgl-recovery.spec.ts 1 test-no-polling-loop:tests/e2e/floating-workspace-shared-glyph-atlas.spec.ts -1 test-no-polling-loop:tests/e2e/golden-core-flows.spec.ts 1 test-no-polling-loop:tests/e2e/helpers/computer-cli-driver.ts 2 test-no-polling-loop:tests/e2e/helpers/computer-driver.ts 1 test-no-polling-loop:tests/e2e/helpers/daemon-generation-processes.ts @@ -302,7 +301,6 @@ 4 test-no-real-time-wait:tests/e2e/floating-workspace-reopen-webgl-recovery.spec.ts 4 test-no-real-time-wait:tests/e2e/floating-workspace-shared-glyph-atlas.spec.ts 2 test-no-real-time-wait:tests/e2e/github-project-board.spec.ts -1 test-no-real-time-wait:tests/e2e/golden-core-flows.spec.ts 1 test-no-real-time-wait:tests/e2e/headless-paired-remote-terminal-reveal-grid.spec.ts 1 test-no-real-time-wait:tests/e2e/headless-paired-remote-terminal-viewer-size.spec.ts 2 test-no-real-time-wait:tests/e2e/helpers/computer-cli-driver.ts @@ -895,7 +893,6 @@ 2 test-timeout-needs-reason:tests/e2e/sidebar-unread-older-paired-host.spec.ts 1 test-timeout-needs-reason:tests/e2e/sidebar-workspace-rows.spec.ts 1 test-timeout-needs-reason:tests/e2e/source-control-large-file-count.spec.ts -5 test-timeout-needs-reason:tests/e2e/ssh-orcad-auto-convert.spec.ts 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-browser-drop-owner.spec.ts 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-browser-routing.spec.ts 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-browser-service-status.spec.ts @@ -925,7 +922,7 @@ 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-terminal-root-owner.spec.ts 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-unbindable-tunnel-port.spec.ts 1 test-timeout-needs-reason:tests/e2e/ssh-orcad-update-anyway.spec.ts -2 test-timeout-needs-reason:tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts +1 test-timeout-needs-reason:tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts 4 test-timeout-needs-reason:tests/e2e/ssh-orcad-workspace-tools.spec.ts 1 test-timeout-needs-reason:tests/e2e/staging-skill-sharing.spec.ts 1 test-timeout-needs-reason:tests/e2e/structured-chat-tabs-open-close.spec.ts diff --git a/config/reliability-gates.jsonc b/config/reliability-gates.jsonc index 2872dd00dc7..98d6578a5bd 100644 --- a/config/reliability-gates.jsonc +++ b/config/reliability-gates.jsonc @@ -1354,10 +1354,10 @@ "providers": ["local", "ssh", "remote-runtime", "mobile"], "coveredPlatforms": [], "coveredProviders": [], - "coverageNotes": "Audit1659 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1474 removes the full renderer session-hook suite. Supplied-state projection, mock transport/settings forwarding and assembled shared-owner policy alarms are retired; no equivalent replacement coverage is claimed. The option admission assertion reference was already stale. Reduced selectors have no fresh focused run; historical execution evidence retains its original scope. Audit1308 changes referenced sources. Retained unchanged-source receipts apply only to their unchanged commands; the reduced selectors have no new execution result. Historical coverage account (not a current claim for changed sources): Deterministic host and direct/Relay transport tests cover per-operation duplicate protection, caller changes, missing host journal rows, acknowledgement loss, and expiry. Mobile tests cover new-action identity, remount, storage refusal, current attachment paths, and capability changes; renderer tests cover one request per send and the bounded hand-back. Live SSH/remote runtimes and physical iOS/Android lifecycle are not exercised. The test-value audit removes the mock send-bridge suite: skills/options/attachment forwarding and accepted/unknown/queued/rejected callback policy no longer have that isolated alarm. Remaining send/draft tests are not claimed as equivalent coverage. Historical run commands and counts are preserved as historical records; a narrowed selector is not a new passed run. Audit1294 removes mobile/src/session/use-mobile-structured-agent-session-newer-approval-send.test.tsx. Its mock-only request/capability/private-state/copy alarms are intentionally lost. Remaining tests are not equivalent replacement coverage. Narrowed commands require fresh parent-owned validation; no new runtime, platform or physical mobile evidence is claimed. Audit1302: src/main/native-chat/agent-session-wire/structured-agent-session-send-idempotency.test.ts: Direct performSend composition seeds real journal outcomes but adapter dispatch is never reachable from performSend at all (owner lines90–159: delivery runs separately). Thus both never-redispatch claims are unexercised; caller ledgers are not constructed and supplied acceptance callbacks never execute. Remaining assertions copy seeded unknown state or count one queued row. Lose isolated submission replay/dedup characterization; no equivalent replacement. Reduced selectors require fresh validation; prior full-suite receipts do not validate removed or changed coverage. No equivalent replacement or physical-provider claim. Audit1303 intentionally removes supplied-state/mock-callback component and hook examples in mobile/src/session/mobile-structured-send-delivery.test.ts, mobile/src/session/use-mobile-structured-agent-session-background-tasks.test.tsx. Their exact retired alarms are not equivalent to retained coverage; no new runtime, physical phone or deployed mixed-version evidence is claimed. Audit1308: src/renderer/src/lib/launch-structured-agent-session.test.ts removed. Unresolvable probe retries, exhausted selector refusal and unknown-outcome create-envelope mock alarms disappear. No equivalent replacement coverage is claimed. Audit1315 removes/narrows mobile/src/session/use-mobile-structured-agent-session-prompt-cancel.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands. Audit1341 scope1334 removes/narrows mobile/src/mobile-web-shell/use-page-host-snapshot.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands. Audit1341 scope1335 removes/narrows mobile/src/session/native-chat/composer/use-mobile-native-chat-image-attachments.test.ts, mobile/src/session/use-mobile-structured-agent-session-send.test.tsx, mobile/src/session/use-mobile-structured-agent-session.test.tsx, mobile/src/transport/mobile-relay-rpc-session.test.ts, mobile/src/transport/stable-logical-rpc-client.test.ts, mobile/src/session/use-mobile-structured-agent-session-queued.test.tsx, mobile/src/mobile-web-shell/bridge-host-init.test.ts. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands.", + "coverageNotes": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Retire the full mobile structured-session hook and launch suites, including their mocked send/abort state projections. Current commands, testFiles and evidenceRuns are already empty; prior narratives and archived receipts retain original scopes. Audit1659 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1474 removes the full renderer session-hook suite. Supplied-state projection, mock transport/settings forwarding and assembled shared-owner policy alarms are retired; no equivalent replacement coverage is claimed. The option admission assertion reference was already stale. Reduced selectors have no fresh focused run; historical execution evidence retains its original scope. Audit1308 changes referenced sources. Retained unchanged-source receipts apply only to their unchanged commands; the reduced selectors have no new execution result. Historical coverage account (not a current claim for changed sources): Deterministic host and direct/Relay transport tests cover per-operation duplicate protection, caller changes, missing host journal rows, acknowledgement loss, and expiry. Mobile tests cover new-action identity, remount, storage refusal, current attachment paths, and capability changes; renderer tests cover one request per send and the bounded hand-back. Live SSH/remote runtimes and physical iOS/Android lifecycle are not exercised. The test-value audit removes the mock send-bridge suite: skills/options/attachment forwarding and accepted/unknown/queued/rejected callback policy no longer have that isolated alarm. Remaining send/draft tests are not claimed as equivalent coverage. Historical run commands and counts are preserved as historical records; a narrowed selector is not a new passed run. Audit1294 removes mobile/src/session/use-mobile-structured-agent-session-newer-approval-send.test.tsx. Its mock-only request/capability/private-state/copy alarms are intentionally lost. Remaining tests are not equivalent replacement coverage. Narrowed commands require fresh parent-owned validation; no new runtime, platform or physical mobile evidence is claimed. Audit1302: src/main/native-chat/agent-session-wire/structured-agent-session-send-idempotency.test.ts: Direct performSend composition seeds real journal outcomes but adapter dispatch is never reachable from performSend at all (owner lines90–159: delivery runs separately). Thus both never-redispatch claims are unexercised; caller ledgers are not constructed and supplied acceptance callbacks never execute. Remaining assertions copy seeded unknown state or count one queued row. Lose isolated submission replay/dedup characterization; no equivalent replacement. Reduced selectors require fresh validation; prior full-suite receipts do not validate removed or changed coverage. No equivalent replacement or physical-provider claim. Audit1303 intentionally removes supplied-state/mock-callback component and hook examples in mobile/src/session/mobile-structured-send-delivery.test.ts, mobile/src/session/use-mobile-structured-agent-session-background-tasks.test.tsx. Their exact retired alarms are not equivalent to retained coverage; no new runtime, physical phone or deployed mixed-version evidence is claimed. Audit1308: src/renderer/src/lib/launch-structured-agent-session.test.ts removed. Unresolvable probe retries, exhausted selector refusal and unknown-outcome create-envelope mock alarms disappear. No equivalent replacement coverage is claimed. Audit1315 removes/narrows mobile/src/session/use-mobile-structured-agent-session-prompt-cancel.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands. Audit1341 scope1334 removes/narrows mobile/src/mobile-web-shell/use-page-host-snapshot.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands. Audit1341 scope1335 removes/narrows mobile/src/session/native-chat/composer/use-mobile-native-chat-image-attachments.test.ts, mobile/src/session/use-mobile-structured-agent-session-send.test.tsx, mobile/src/session/use-mobile-structured-agent-session.test.tsx, mobile/src/transport/mobile-relay-rpc-session.test.ts, mobile/src/transport/stable-logical-rpc-client.test.ts, mobile/src/session/use-mobile-structured-agent-session-queued.test.tsx, mobile/src/mobile-web-shell/bridge-host-init.test.ts. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands.", "motivatingLinks": ["https://github.com/stablyai/orca/pull/20133"], "invariant": "One structured-send operation id causes at most one provider dispatch. Each user Send on phone or desktop is a new action with a fresh id and one request; neither automatically resends an ambiguous action. The host retains the original action's outcome, and a later phone Send remains possible with equal text, after remount, or when phone storage fails. Repeated transmission of the same operation remains protected by host deduplication.", - "oracle": "The retained unchanged-source receipts are the only current execution evidence; reduced-source assertions await validation. Historical oracle account (not a current claim for changed sources): Inject adapter/RPC acknowledgement loss, caller replacement, logical-client close after response, auth recovery with a written request, missing host submissions, legacy pending rows, stale fences, expiry, mobile remount, and phone storage refusal. Assert at most one dispatch for repeated transmission of one operation. For a later phone Send, assert a fresh id and second delivery while the original host entry remains unknown; equal images use current uploaded paths and delivery fields follow current host capabilities.", + "oracle": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Retire the full mobile structured-session hook and launch suites, including their mocked send/abort state projections. Current commands, testFiles and evidenceRuns are already empty; prior narratives and archived receipts retain original scopes. The retained unchanged-source receipts are the only current execution evidence; reduced-source assertions await validation. Historical oracle account (not a current claim for changed sources): Inject adapter/RPC acknowledgement loss, caller replacement, logical-client close after response, auth recovery with a written request, missing host submissions, legacy pending rows, stale fences, expiry, mobile remount, and phone storage refusal. Assert at most one dispatch for repeated transmission of one operation. For a later phone Send, assert a fresh id and second delivery while the original host entry remains unknown; equal images use current uploaded paths and delivery fields follow current host capabilities.", "commands": [], "testFiles": [], "assertionRefs": [], @@ -1406,7 +1406,8 @@ "Retired during audit1312 selector cleanup; original complete execution receipt (unchanged): {\"date\":\"2026-11-04\",\"runner\":\"local\",\"platform\":\"linux\",\"result\":\"passed\",\"command\":\"ORCA_BACKGROUND_LAUNCH=1 pnpm --config.verifyDepsBeforeRun=false test src/shared/agent-session/lifecycle/agent-session-operation-ledger.test.ts src/shared/agent-session/requests/structured-agent-session-send-evidence.test.ts src/main/runtime/agent-session/ownership/agent-session-operation-admission.test.ts src/main/native-chat/agent-session-wire/structured-agent-session-send.test.ts src/main/runtime/orchestration/structured-mailbox-pointer-delivery.test.ts src/main/runtime/orchestration/structured-mailbox-pointer-host.test.ts src/renderer/src/components/native-chat/structured-agent-session-message-sender.test.ts src/renderer/src/components/native-chat/use-structured-agent-session.test.tsx src/renderer/src/lib/launch-structured-agent-session.test.ts --maxWorkers=2\",\"currentCommand\":\"ORCA_BACKGROUND_LAUNCH=1 pnpm test src/shared/agent-session/lifecycle/agent-session-operation-ledger.test.ts src/shared/agent-session/requests/structured-agent-session-send-evidence.test.ts src/main/runtime/agent-session/ownership/agent-session-operation-admission.test.ts src/main/native-chat/agent-session-wire/structured-agent-session-send.test.ts src/main/runtime/orchestration/structured-mailbox-pointer-delivery.test.ts src/main/runtime/orchestration/structured-mailbox-pointer-host.test.ts src/renderer/src/components/native-chat/structured-agent-session-message-sender.test.ts src/renderer/src/components/native-chat/use-structured-agent-session.test.tsx src/renderer/src/lib/launch-structured-agent-session.test.ts\",\"durationSeconds\":11.86,\"summary\":\"Test Files 9 passed (9); Tests 145 passed (145). Focused unit selector; native adapters remain mocked where present. Single local execution, not a p95 or cross-platform measurement. Source head 36536504876bf7fa79c79256377cde97f1f3a133.\"}", "Audit1315 removes/narrows mobile/src/session/use-mobile-structured-agent-session-prompt-cancel.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands.", "Audit1341 scope1334 removes/narrows mobile/src/mobile-web-shell/use-page-host-snapshot.test.tsx. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands.", - "Audit1341 scope1335 removes/narrows mobile/src/session/native-chat/composer/use-mobile-native-chat-image-attachments.test.ts, mobile/src/session/use-mobile-structured-agent-session-send.test.tsx, mobile/src/session/use-mobile-structured-agent-session.test.tsx, mobile/src/transport/mobile-relay-rpc-session.test.ts, mobile/src/transport/stable-logical-rpc-client.test.ts, mobile/src/session/use-mobile-structured-agent-session-queued.test.tsx, mobile/src/mobile-web-shell/bridge-host-init.test.ts. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands." + "Audit1341 scope1335 removes/narrows mobile/src/session/native-chat/composer/use-mobile-native-chat-image-attachments.test.ts, mobile/src/session/use-mobile-structured-agent-session-send.test.tsx, mobile/src/session/use-mobile-structured-agent-session.test.tsx, mobile/src/transport/mobile-relay-rpc-session.test.ts, mobile/src/transport/stable-logical-rpc-client.test.ts, mobile/src/session/use-mobile-structured-agent-session-queued.test.tsx, mobile/src/mobile-web-shell/bridge-host-init.test.ts. Exact retired scenarios are intentionally lost; surviving assertions are limited to unchanged retained authored cases. No equivalent replacement, physical phone or deployed mixed-host claim. Fresh validation is required for narrowed commands.", + "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Retire the full mobile structured-session hook and launch suites, including their mocked send/abort state projections. Current commands, testFiles and evidenceRuns are already empty; prior narratives and archived receipts retain original scopes. All prior notes and dated/archived evidence retain original scopes." ], "demotionRule": "Keep experimental or demote if one operation dispatches twice, uncertainty is reported as definite rejection, a later phone Send reuses the earlier action or is blocked by phone storage, or the focused commands flake without a diagnosed harness defect." }, @@ -4075,10 +4076,10 @@ "providers": ["remote-runtime"], "coveredPlatforms": ["macos"], "coveredProviders": ["remote-runtime"], - "coverageNotes": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit1640 removes low-value IPC, SSH, workspace and transcript-watcher tests without equivalent replacement coverage. Per-source reviews record retained public refusal, cancellation, execution-host and real-filesystem outcomes with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Deterministic main-IPC contract tests cover disconnect-driven close delivery, exactly-once close, per-subscription teardown isolation against a failing socket close and a throwing liveness probe, containment of a throwing renderer send on the unguarded host-close path, and continued suppression of stale payloads from a retired transport. A headed paired-server journey (real Orca host plus a separate paired Orca desktop client) covers hidden-but-mounted reveal, cold-parked reveal, and cold-parked reveal across a disconnect/reconnect. Live Linux and Windows paired-server evidence and real sleep/wake transport loss remain uncollected.", + "coverageNotes": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit1640 removes low-value IPC, SSH, workspace and transcript-watcher tests without equivalent replacement coverage. Per-source reviews record retained public refusal, cancellation, execution-host and real-filesystem outcomes with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Deterministic main-IPC contract tests cover disconnect-driven close delivery, exactly-once close, per-subscription teardown isolation against a failing socket close and a throwing liveness probe, containment of a throwing renderer send on the unguarded host-close path, and continued suppression of stale payloads from a retired transport. A headed paired-server journey (real Orca host plus a separate paired Orca desktop client) covers hidden-but-mounted reveal, cold-parked reveal, and cold-parked reveal across a disconnect/reconnect. Live Linux and Windows paired-server evidence and real sleep/wake transport loss remain uncollected.", "motivatingLinks": ["tests/e2e/paired-remote-terminal-parked-reveal-interactivity.spec.ts"], "invariant": "Every renderer-held runtime subscription receives exactly one terminal close event when its transport is retired, including when the retirement advanced the transport generation first, and a single failing teardown never abandons that environment's remaining subscriptions nor escapes into the transport that reported the close. Payload frames from a retired transport stay suppressed. A revealed remote terminal therefore reattaches over a live multiplex connection: its buffer restores, typed input reaches the host PTY, the echo paints without a tab flip, and the PTY converges on the revealed pane grid.", - "oracle": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): The main IPC contract test subscribes terminal.multiplex through the real handler, disconnects the environment, and asserts the renderer received exactly one {type: close} subscription event. Two isolation tests subscribe a second stream to the same environment and make the first one fail -- in its socket close, and in the liveness probe inside notifyClosed -- then assert the disconnect does not throw, both transports closed, and every close the renderer could still receive was delivered. A third drives a host-initiated close through the transport callback, which is the one notifyClosed call site with no surrounding guard, with a renderer send that throws, and asserts it cannot escape into the WebSocket close handler. A fourth test asserts that after retirement a late response frame is not forwarded and a late transport close does not re-send. The paired-server journey runs three reveal scenarios against one real host and one real paired desktop client, and for each records buffer restore, host-side receipt of the typed marker through an out-of-band host sink file, live paint without a tab flip, and PTY-versus-pane grid convergence.", + "oracle": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): The main IPC contract test subscribes terminal.multiplex through the real handler, disconnects the environment, and asserts the renderer received exactly one {type: close} subscription event. Two isolation tests subscribe a second stream to the same environment and make the first one fail -- in its socket close, and in the liveness probe inside notifyClosed -- then assert the disconnect does not throw, both transports closed, and every close the renderer could still receive was delivered. A third drives a host-initiated close through the transport callback, which is the one notifyClosed call site with no surrounding guard, with a renderer send that throws, and asserts it cannot escape into the WebSocket close handler. A fourth test asserts that after retirement a late response frame is not forwarded and a late transport close does not re-send. The paired-server journey runs three reveal scenarios against one real host and one real paired desktop client, and for each records buffer restore, host-side receipt of the typed marker through an out-of-band host sink file, live paint without a tab flip, and PTY-versus-pane grid convergence.", "commands": [ "pnpm exec vitest run --config config/vitest.config.ts", "pnpm exec playwright test tests/e2e/paired-remote-terminal-parked-reveal-interactivity.spec.ts --config tests/playwright.config.ts --project electron-headless --workers=1" @@ -4207,7 +4208,9 @@ "Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. All prior notes, assertion references and complete dated receipts retain their original historical scopes.", "Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. All prior notes and archived receipts retain original scopes.", "Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", - "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." + "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", + "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. All prior notes and dated/archived evidence retain original scopes.", + "Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." ], "demotionRule": "Demote if a retired runtime transport can leave a renderer subscription without a close event, if a close is delivered more than once, if one failing teardown strands its sibling subscriptions, if payload frames from a retired transport reach the renderer, or if a revealed remote terminal can stay blank or reject input after a reconnect." }, @@ -8876,7 +8879,7 @@ "providers": ["local", "daemon", "ssh", "remote-runtime"], "coveredPlatforms": ["macos"], "coveredProviders": [], - "coverageNotes": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Module layout SSH policy: Retire the multiplexer private-clock saturation registration, the writer saturation transition-count registration and the changed-source dated writer/backpressure receipt. Retained byte delivery, lane order, fairness, settlement and cancellation cases are unvalidated by that historical receipt. No equivalent replacement, new execution, measured performance, live SSH/WSL or cross-platform proof is claimed. The unchanged WSL protocol receipt retains only its original command scope. Historical prior account (preserved verbatim): Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2070 (sources2068/2069): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/hooks-and-hosted-review.spec'). src/main/ssh/ssh-multiplexer-transport-writer.test.ts: reports saturated epochs and their drain boundary exactly once. src/main/runtime/orca-runtime-tests/worktree-removal-archive-hook-gate.spec.ts: removes and records the waiver when the failure is explicitly overridden; removes without an override record when the hook succeeds; removes when the hook is configured but not requested; removes when no archive hook is configured; holds a desktop removal in the same repo until this one is accepted; joins a desktop removal of the same worktree accepted while this one waited. src/main/runtime/orca-runtime-tests/worktree-ps-agent-row-dismissal.spec.ts: drops the row as soon as the user dismisses it, without waiting for the PTY to exit; tells paired clients to republish on the transition and on the dismissal; rejoins a row through its %s handle after pane ownership clears. src/main/runtime/orca-runtime-tests/hooks-and-hosted-review.spec.ts: reports an error when the SSH filesystem provider is unavailable; reports an SSH orca.yaml as an error without reading any filesystem; reports ok for a local repo hook check; reads SSH issue commands as absent and drops writes without touching any filesystem; allows host integration slug helpers for SSH repos through provider-aware GitHub clients; routes runtime GitHub repo identity helpers through the selected WSL project runtime; routes runtime GitHub issue and work-item actions through the selected WSL project runtime; pins explicit origin preference on runtime open-by-number work item lookups; forwards check-details cancellation without local Git overrides. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. The historical receipt/currentCommand selecting src/main/ssh/ssh-multiplexer-transport-writer.test.ts remains historical and is not a passing result for the reduced current command. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Delete the complete tail-wait memo suite: its sole current authored callback checks supplied fromTail/blocked field values. Retire all three broader memoization/equivalence/eviction assertion aliases; they are not backed by retained source callbacks. No scan-work budget, sustained output, physical terminal or resource measurement is assigned to this removed test. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1674 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1634 removes low-value IPC, filesystem and runtime projections, argument inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained cancellation, ownership, authorization, filesystem and SQLite boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Local macOS evidence covers main-process pending-output and projection-admission caps, saturated-tail wait detection, direct-SSH source retention/accounting, always-on V1 negotiation with legacy peer fallback, exact provider-generation pause ownership, renderer exit/data ordering, decoder input bounds, bounded reconnect scheduling, and WSL stdio transport settlement through deterministic unit contracts. A joined main/runtime oracle exercises renderer-sourced headed semantics and headless-model snapshot semantics without launching a live paired server. Separately, a macOS-hosted Docker OpenSSH run exercises only the deployed Linux relay and direct SSH provider with an exact 256 KiB source-credit plateau, concurrent PTY typing, fixed-size filesystem frames, Git churn, and owner-lease reconnect. Neither deterministic topology labels nor Docker SSH constitute live headed desktop, headless orca serve, or physical WSL evidence; local/daemon, prior-version daemons, Windows named pipes/ConPTY, folder workspaces, and mixed-version clients also remain uncovered. Audit1312 removes low-value imported specs from aggregate test drivers; old whole-driver execution receipts are archived. Retained registrations are unchanged. No equivalent replacement coverage is claimed. Audit1383: src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts reduced. Lose command-list/receipt/default/background activation matrices whose provider spawn returns canned IDs. Retained test proves actual tab allocation separation, not terminal startup. No equivalent replacement coverage is claimed. Existing exact unaffected receipts and their platform/provider/budget claims remain scoped to unchanged selectors. Changed optional selectors have no new execution result. Audit1401: src/main/runtime/orca-runtime-tests/mobile-session-tabs-duplicate-pty-repair.spec.ts removed. Loses listing-on-already-repaired-data alarm; it never established runtime repair invocation. src/main/runtime/orca-runtime-tests/mobile-session-tabs-split-agent-tab.spec.ts removed. Loses actual mobile split sibling shell versus recorded-agent resume command alarms; no equivalent coverage asserted. src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts removed. Loses separate setup/startup tab result in that manually assembled path. src/main/runtime/orca-runtime.test.ts reduced. Driver registration/import of a whole-deleted selected spec is removed; directly authored cases remain unchanged. No extra authored removal credit. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result.", + "coverageNotes": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Module layout SSH policy: Retire the multiplexer private-clock saturation registration, the writer saturation transition-count registration and the changed-source dated writer/backpressure receipt. Retained byte delivery, lane order, fairness, settlement and cancellation cases are unvalidated by that historical receipt. No equivalent replacement, new execution, measured performance, live SSH/WSL or cross-platform proof is claimed. The unchanged WSL protocol receipt retains only its original command scope. Historical prior account (preserved verbatim): Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2070 (sources2068/2069): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/hooks-and-hosted-review.spec'). src/main/ssh/ssh-multiplexer-transport-writer.test.ts: reports saturated epochs and their drain boundary exactly once. src/main/runtime/orca-runtime-tests/worktree-removal-archive-hook-gate.spec.ts: removes and records the waiver when the failure is explicitly overridden; removes without an override record when the hook succeeds; removes when the hook is configured but not requested; removes when no archive hook is configured; holds a desktop removal in the same repo until this one is accepted; joins a desktop removal of the same worktree accepted while this one waited. src/main/runtime/orca-runtime-tests/worktree-ps-agent-row-dismissal.spec.ts: drops the row as soon as the user dismisses it, without waiting for the PTY to exit; tells paired clients to republish on the transition and on the dismissal; rejoins a row through its %s handle after pane ownership clears. src/main/runtime/orca-runtime-tests/hooks-and-hosted-review.spec.ts: reports an error when the SSH filesystem provider is unavailable; reports an SSH orca.yaml as an error without reading any filesystem; reports ok for a local repo hook check; reads SSH issue commands as absent and drops writes without touching any filesystem; allows host integration slug helpers for SSH repos through provider-aware GitHub clients; routes runtime GitHub repo identity helpers through the selected WSL project runtime; routes runtime GitHub issue and work-item actions through the selected WSL project runtime; pins explicit origin preference on runtime open-by-number work item lookups; forwards check-details cancellation without local Git overrides. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. The historical receipt/currentCommand selecting src/main/ssh/ssh-multiplexer-transport-writer.test.ts remains historical and is not a passing result for the reduced current command. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Delete the complete tail-wait memo suite: its sole current authored callback checks supplied fromTail/blocked field values. Retire all three broader memoization/equivalence/eviction assertion aliases; they are not backed by retained source callbacks. No scan-work budget, sustained output, physical terminal or resource measurement is assigned to this removed test. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1674 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1634 removes low-value IPC, filesystem and runtime projections, argument inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained cancellation, ownership, authorization, filesystem and SQLite boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Local macOS evidence covers main-process pending-output and projection-admission caps, saturated-tail wait detection, direct-SSH source retention/accounting, always-on V1 negotiation with legacy peer fallback, exact provider-generation pause ownership, renderer exit/data ordering, decoder input bounds, bounded reconnect scheduling, and WSL stdio transport settlement through deterministic unit contracts. A joined main/runtime oracle exercises renderer-sourced headed semantics and headless-model snapshot semantics without launching a live paired server. Separately, a macOS-hosted Docker OpenSSH run exercises only the deployed Linux relay and direct SSH provider with an exact 256 KiB source-credit plateau, concurrent PTY typing, fixed-size filesystem frames, Git churn, and owner-lease reconnect. Neither deterministic topology labels nor Docker SSH constitute live headed desktop, headless orca serve, or physical WSL evidence; local/daemon, prior-version daemons, Windows named pipes/ConPTY, folder workspaces, and mixed-version clients also remain uncovered. Audit1312 removes low-value imported specs from aggregate test drivers; old whole-driver execution receipts are archived. Retained registrations are unchanged. No equivalent replacement coverage is claimed. Audit1383: src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts reduced. Lose command-list/receipt/default/background activation matrices whose provider spawn returns canned IDs. Retained test proves actual tab allocation separation, not terminal startup. No equivalent replacement coverage is claimed. Existing exact unaffected receipts and their platform/provider/budget claims remain scoped to unchanged selectors. Changed optional selectors have no new execution result. Audit1401: src/main/runtime/orca-runtime-tests/mobile-session-tabs-duplicate-pty-repair.spec.ts removed. Loses listing-on-already-repaired-data alarm; it never established runtime repair invocation. src/main/runtime/orca-runtime-tests/mobile-session-tabs-split-agent-tab.spec.ts removed. Loses actual mobile split sibling shell versus recorded-agent resume command alarms; no equivalent coverage asserted. src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts removed. Loses separate setup/startup tab result in that manually assembled path. src/main/runtime/orca-runtime.test.ts reduced. Driver registration/import of a whole-deleted selected spec is removed; directly authored cases remain unchanged. No extra authored removal credit. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result.", "motivatingLinks": [ "https://github.com/stablyai/orca/pull/6836", "https://github.com/stablyai/orca/pull/6858", @@ -8884,7 +8887,7 @@ "https://github.com/stablyai/orca/pull/7054" ], "invariant": "High-volume terminal output must stay bounded across daemon socket writes, SSH relay writer/decoder/source retention/reconnect, main model and projection admission, runtime metadata, detectors, and tail checks, main-to-renderer in-flight bytes, renderer scheduler queues, and hidden-output restore without starving focused input. Every new SSH session offers V1 and every same-build relay supports it; legacy delivery is reachable only when capability negotiation proves the peer cannot use V1. Negotiated SSH spans remain exact and contiguous through activation, exit, cancellation, recovery, desktop projection, and required remote replacement; provisional activation data cannot project, reconnect may release it only into the exact attempt's private recovery quarantine after contract validation, and exit seals that quarantine against later same-token frames until ordered admission or exact cancellation proof. Reconnect checkpoint capture freezes exact provider-generation/PTy admission, cancels queued old work, and waits for the running raw completion before exposing its accepted checkpoint; timeout or failure detaches only that PTy's old model, makes the checkpoint unavailable, and releases the bounded fence without closing the shared provider, while the same failure outside an active exact migration remains generation-fatal and overlapping reconnects preserve any earlier outstanding fence. Pending renderer projection IDs are capped and compacted across split remainders, exit preparation owns its renderer fence through finalization, exit-time cancellation transfers published projections before proof commit, generation close fences late proof while draining exact projection waiters, closed-generation identity compacts without weakening stale rejection, canceled source-delivery retirement retains at most one ordered token per PTY, and pause/resume targets only the exact provider generation. Additional subscribers cannot stall the owner, while the required legacy primary retains backpressure.", - "oracle": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Module layout SSH policy: Retire the multiplexer private-clock saturation registration, the writer saturation transition-count registration and the changed-source dated writer/backpressure receipt. Retained byte delivery, lane order, fairness, settlement and cancellation cases are unvalidated by that historical receipt. No equivalent replacement, new execution, measured performance, live SSH/WSL or cross-platform proof is claimed. The unchanged WSL protocol receipt retains only its original command scope. Historical prior account (preserved verbatim): Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2070 (sources2068/2069): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/hooks-and-hosted-review.spec'). src/main/ssh/ssh-multiplexer-transport-writer.test.ts: reports saturated epochs and their drain boundary exactly once. src/main/runtime/orca-runtime-tests/worktree-removal-archive-hook-gate.spec.ts: removes and records the waiver when the failure is explicitly overridden; removes without an override record when the hook succeeds; removes when the hook is configured but not requested; removes when no archive hook is configured; holds a desktop removal in the same repo until this one is accepted; joins a desktop removal of the same worktree accepted while this one waited. src/main/runtime/orca-runtime-tests/worktree-ps-agent-row-dismissal.spec.ts: drops the row as soon as the user dismisses it, without waiting for the PTY to exit; tells paired clients to republish on the transition and on the dismissal; rejoins a row through its %s handle after pane ownership clears. src/main/runtime/orca-runtime-tests/hooks-and-hosted-review.spec.ts: reports an error when the SSH filesystem provider is unavailable; reports an SSH orca.yaml as an error without reading any filesystem; reports ok for a local repo hook check; reads SSH issue commands as absent and drops writes without touching any filesystem; allows host integration slug helpers for SSH repos through provider-aware GitHub clients; routes runtime GitHub repo identity helpers through the selected WSL project runtime; routes runtime GitHub issue and work-item actions through the selected WSL project runtime; pins explicit origin preference on runtime open-by-number work item lookups; forwards check-details cancellation without local Git overrides. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. The historical receipt/currentCommand selecting src/main/ssh/ssh-multiplexer-transport-writer.test.ts remains historical and is not a passing result for the reduced current command. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Delete the complete tail-wait memo suite: its sole current authored callback checks supplied fromTail/blocked field values. Retire all three broader memoization/equivalence/eviction assertion aliases; they are not backed by retained source callbacks. No scan-work budget, sustained output, physical terminal or resource measurement is assigned to this removed test. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1674 retires changed-source assertion and execution scope without equivalent replacement coverage. Existing unchanged-selector receipts remain scoped to their original sources. Historical prior-source oracle: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: The current executable slice injects main-process renderer backlog pressure, then asserts ordinary terminal-wait detection without a joined-tail allocation and with full prompt-family continuity, per-PTY and total pending-output caps, a 1,024-ID projection cap with split-remainder compaction, preserved sequenced-tail metadata, active-pending protection ahead of background trimming, exact provider-generation pause ownership, renderer exit/data ordering, and ACK-gated in-flight bounds. The negotiation seam proves every initial connection and reconnect offers V1, the same-build relay advertises V1 without launch flags, old clients remain token-free, and method-not-found peers fall back without installing source ACK publication. The joined source-intake/multiplex oracle blocks renderer or headless serialization, admits one snapshot-covered span and one trailing span, requires replacement reservation only after the authoritative sequence is known, advances upstream ACK eligibility exactly to the covered boundary after SnapshotEnd, and delivers then ACKs the trailing span through the ordinary live path without cancellation. Source-range ledger and multiplex seams additionally require cumulative byte ACK 40 then 100 for one 100-byte frame to release exact byte credit without early source settlement, admit a contiguous higher-generation recovered token while the prior token remains unsettled, and reject stale stream or source generations without releasing in-flight byte credit. Replacement reservation rolls back every earlier span on a later failure, commit requires the exact frozen transfer states, and an authoritatively reclaimed covered span rejects commit without local trim authority while rollback removes the reservation idempotently. The exit-deadline oracle publishes a source-backed desktop projection, advances a fake clock to cancellation, and requires one projection transfer, proof commit, preparation, and final exit with no provider close, retained obligation, duplicate finalization, or process-lifetime cancellation tombstone; a second controlled proof remains pending across generation close and must never publish final exit, while both paths release the renderer preparation lease exactly once. The model-migration oracle accepts span A, blocks span B's raw emulator completion, begins disconnect migration, and proves no checkpoint or reconnect attach can remain at A; releasing B advances the exact checkpoint to B and projects B once. Its fake-clock arm reaches the 10-second deadline, requires checkpoint-unavailable, one exact model reset, zero retained admission charge and timers, and no effect from late raw settlement. Its two-PTY failure arm rejects one running raw callback under an active exact migration and requires one checkpoint-unavailable result, one model reset, zero provider closes, retained sibling checkpoint and transport liveness, zero charges and timers, and an unaffected other generation; a non-migrating sibling oracle still requires generation close. The intake additionally closes 2,048 sequential provider generations into one exact range, rejects stale events across that range, and preserves an out-of-order live gap until its own close. Direct-SSH contracts additionally prove activation-response settlement before the first source frame, claim-gated projection under same-decoder-turn response/data delivery, exact early-ACK reservation through send settlement or same-token retry, recovery completion visible before lease-held frames transfers those frames only into the private quarantine and admits no output until exact checkpoint-to-recoveryEnd coverage, invalid-checkpoint restore retention through response settlement followed by fresh-token retry and one live source frame, isolated saturated-subscriber eviction with healthy owner liveness, preserved legacy-primary backpressure, one-write ownership across write(false)/drain, control selection ahead of queued ordinary PTY input with FIFO lanes and bounded fairness, liveness rebasing, failed-exit retention through late ACK or exact recovery, response-settlement cancellation authority, token-scoped exit cancellation, exact checkpoint-to-recoveryEnd continuity, stale-owner one-shot retry, no physical PTY teardown on recovery failure, exact provider-generation closure and once-only cleanup when recovery cancellation publication or proof rejects, 16 MiB-plus-1 MiB decoder caps, charged per-PTY/session retention budgets, and eight-wide isolated reattach. The live Electron perf oracle adds hidden-output floods, renderer scheduler queue depth, dropped-output-zero normal scenarios, and active key latency budgets before promotion.", + "oracle": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Module layout SSH policy: Retire the multiplexer private-clock saturation registration, the writer saturation transition-count registration and the changed-source dated writer/backpressure receipt. Retained byte delivery, lane order, fairness, settlement and cancellation cases are unvalidated by that historical receipt. No equivalent replacement, new execution, measured performance, live SSH/WSL or cross-platform proof is claimed. The unchanged WSL protocol receipt retains only its original command scope. Historical prior account (preserved verbatim): Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2070 (sources2068/2069): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/hooks-and-hosted-review.spec'). src/main/ssh/ssh-multiplexer-transport-writer.test.ts: reports saturated epochs and their drain boundary exactly once. src/main/runtime/orca-runtime-tests/worktree-removal-archive-hook-gate.spec.ts: removes and records the waiver when the failure is explicitly overridden; removes without an override record when the hook succeeds; removes when the hook is configured but not requested; removes when no archive hook is configured; holds a desktop removal in the same repo until this one is accepted; joins a desktop removal of the same worktree accepted while this one waited. src/main/runtime/orca-runtime-tests/worktree-ps-agent-row-dismissal.spec.ts: drops the row as soon as the user dismisses it, without waiting for the PTY to exit; tells paired clients to republish on the transition and on the dismissal; rejoins a row through its %s handle after pane ownership clears. src/main/runtime/orca-runtime-tests/hooks-and-hosted-review.spec.ts: reports an error when the SSH filesystem provider is unavailable; reports an SSH orca.yaml as an error without reading any filesystem; reports ok for a local repo hook check; reads SSH issue commands as absent and drops writes without touching any filesystem; allows host integration slug helpers for SSH repos through provider-aware GitHub clients; routes runtime GitHub repo identity helpers through the selected WSL project runtime; routes runtime GitHub issue and work-item actions through the selected WSL project runtime; pins explicit origin preference on runtime open-by-number work item lookups; forwards check-details cancellation without local Git overrides. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. The historical receipt/currentCommand selecting src/main/ssh/ssh-multiplexer-transport-writer.test.ts remains historical and is not a passing result for the reduced current command. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Delete the complete tail-wait memo suite: its sole current authored callback checks supplied fromTail/blocked field values. Retire all three broader memoization/equivalence/eviction assertion aliases; they are not backed by retained source callbacks. No scan-work budget, sustained output, physical terminal or resource measurement is assigned to this removed test. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1674 retires changed-source assertion and execution scope without equivalent replacement coverage. Existing unchanged-selector receipts remain scoped to their original sources. Historical prior-source oracle: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: The current executable slice injects main-process renderer backlog pressure, then asserts ordinary terminal-wait detection without a joined-tail allocation and with full prompt-family continuity, per-PTY and total pending-output caps, a 1,024-ID projection cap with split-remainder compaction, preserved sequenced-tail metadata, active-pending protection ahead of background trimming, exact provider-generation pause ownership, renderer exit/data ordering, and ACK-gated in-flight bounds. The negotiation seam proves every initial connection and reconnect offers V1, the same-build relay advertises V1 without launch flags, old clients remain token-free, and method-not-found peers fall back without installing source ACK publication. The joined source-intake/multiplex oracle blocks renderer or headless serialization, admits one snapshot-covered span and one trailing span, requires replacement reservation only after the authoritative sequence is known, advances upstream ACK eligibility exactly to the covered boundary after SnapshotEnd, and delivers then ACKs the trailing span through the ordinary live path without cancellation. Source-range ledger and multiplex seams additionally require cumulative byte ACK 40 then 100 for one 100-byte frame to release exact byte credit without early source settlement, admit a contiguous higher-generation recovered token while the prior token remains unsettled, and reject stale stream or source generations without releasing in-flight byte credit. Replacement reservation rolls back every earlier span on a later failure, commit requires the exact frozen transfer states, and an authoritatively reclaimed covered span rejects commit without local trim authority while rollback removes the reservation idempotently. The exit-deadline oracle publishes a source-backed desktop projection, advances a fake clock to cancellation, and requires one projection transfer, proof commit, preparation, and final exit with no provider close, retained obligation, duplicate finalization, or process-lifetime cancellation tombstone; a second controlled proof remains pending across generation close and must never publish final exit, while both paths release the renderer preparation lease exactly once. The model-migration oracle accepts span A, blocks span B's raw emulator completion, begins disconnect migration, and proves no checkpoint or reconnect attach can remain at A; releasing B advances the exact checkpoint to B and projects B once. Its fake-clock arm reaches the 10-second deadline, requires checkpoint-unavailable, one exact model reset, zero retained admission charge and timers, and no effect from late raw settlement. Its two-PTY failure arm rejects one running raw callback under an active exact migration and requires one checkpoint-unavailable result, one model reset, zero provider closes, retained sibling checkpoint and transport liveness, zero charges and timers, and an unaffected other generation; a non-migrating sibling oracle still requires generation close. The intake additionally closes 2,048 sequential provider generations into one exact range, rejects stale events across that range, and preserves an out-of-order live gap until its own close. Direct-SSH contracts additionally prove activation-response settlement before the first source frame, claim-gated projection under same-decoder-turn response/data delivery, exact early-ACK reservation through send settlement or same-token retry, recovery completion visible before lease-held frames transfers those frames only into the private quarantine and admits no output until exact checkpoint-to-recoveryEnd coverage, invalid-checkpoint restore retention through response settlement followed by fresh-token retry and one live source frame, isolated saturated-subscriber eviction with healthy owner liveness, preserved legacy-primary backpressure, one-write ownership across write(false)/drain, control selection ahead of queued ordinary PTY input with FIFO lanes and bounded fairness, liveness rebasing, failed-exit retention through late ACK or exact recovery, response-settlement cancellation authority, token-scoped exit cancellation, exact checkpoint-to-recoveryEnd continuity, stale-owner one-shot retry, no physical PTY teardown on recovery failure, exact provider-generation closure and once-only cleanup when recovery cancellation publication or proof rejects, 16 MiB-plus-1 MiB decoder caps, charged per-PTY/session retention budgets, and eight-wide isolated reattach. The live Electron perf oracle adds hidden-output floods, renderer scheduler queue depth, dropped-output-zero normal scenarios, and active key latency budgets before promotion.", "commands": [ "pnpm exec vitest run --config config/vitest.config.ts", "pnpm exec vitest run --config config/vitest.config.ts src/main/runtime/construction/orca-runtime.test.ts", @@ -9070,7 +9073,9 @@ "Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. All prior notes, assertion references and complete dated receipts retain their original historical scopes.", "Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. All prior notes and archived receipts retain original scopes.", "Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", - "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." + "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", + "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. All prior notes and dated/archived evidence retain original scopes.", + "Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." ], "demotionRule": "Cannot promote without metric artifacts and bounded-byte assertions." }, @@ -12351,14 +12356,14 @@ "providers": ["local", "remote-runtime", "ssh"], "coveredPlatforms": ["macos", "linux"], "coveredProviders": ["local", "remote-runtime", "ssh"], - "coverageNotes": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2056 (sources2054/2055): Intentionally retire these low-value alarms without replacement: src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts: delivers remote file changes to listeners for the same worktree. src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/mobile-session-tabs-part-14.spec'). src/main/runtime/orca-runtime-tests/terminal-side-effect-facts-part-03.spec.ts: arms the stale-title timer for a seeded working title; arms the Command Code scrape from the noted spawn command; prefers the tracked title over a stale renderer lastTitle in the hydration seed; writes recent output into the current runtime output map; uses a later title handler replacement with the runtime receiver; calls the current injected listener with the runtime receiver; keeps optional reconciliation lazy after its callback is removed. src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-14.spec.ts: lists them, so a paired client can mirror a migrated editor; closes one by retiring it from the host session; keeps an unsaved draft unless the close is forced; shares aggregate work but leaves a replacement pending promise alone; lets a targeted refresh finish while an aggregate refresh waits; refuses a close when the store lacks %s before the guard; keeps a later missing %s on the original caught failure path. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Exact retired assertionRefs, preserved as historical objects only: {\n \"file\": \"src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts\",\n \"assertions\": [\n \"terminal watcher error and setup failure evict the shared entry before retrying callbacks run\",\n \"an old stream end cannot evict the replacement stream, which still delivers to its listener\",\n \"two same-worktree listeners share a synthetic subscription until the final stop unsubscribes and sends unwatch\"\n ]\n } Historical prior account (preserved verbatim): Audit1973: Delete the standalone mocked watcher cleanup suite and its entire sole-file prospective command. Retire the removed file-command transport-abort alias; preserve the independent cancellation suite and final-overflow/error/end cleanup callback, and13 unrelated receipt objects. Whole original changed dated receipts are retired, never narrowed or transferred. This metadata review claims no fresh execution, equivalent replacement, provider/PTY, live remote/mobile, rendered, cross-platform, soak or measured-budget result. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Partial runtime files-watch suite retains controlled Windows close-event waiting, shared-client survival, late physical-exit ownership/deadline cleanup and shutdown joining. Its combined Windows close/deadline/late-exit alias remains backed by unchanged callbacks. Retire native-error-close, pre-publication setup ownership and failed-delete rearm aliases; associated delegation, failure forwarding, removal routing and recovery examples are lost. These tests inject watch/subscription seams and do not observe an actual native watcher or physical process. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1925 (sources1922/1923): Partial runtime watcher-host suite loses mocked option/event/overflow/resubscription forwarding, supplied teardown-error propagation, aborted-signal/no-call and private ten-thousand-waiter-count inventories. Retire the 10,000-caller bound and failed-recovery-subscriber-ending assertion aliases. Keep same-root owner cancellation, pre-ready replacement, held root-close join, stale-generation recovery delivery and initial failed-root ownership mappings backed by retained real-host callbacks with injected subscription/error/physical-exit promises. No actual native watcher, process termination or resource bound was observed. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1683 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1645 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1640 removes low-value IPC, SSH, workspace and transcript-watcher tests without equivalent replacement coverage. Per-source reviews record retained public refusal, cancellation, execution-host and real-filesystem outcomes with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1634 removes low-value IPC, filesystem and runtime projections, argument inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained cancellation, ownership, authorization, filesystem and SQLite boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1627 removes low-value mocked daemon projections, scalar tables and duplicate terminal examples without equivalent replacement coverage. Per-source reviews record retained terminal content preservation, process ownership refusal and recovery boundaries together with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1623 removes low-value mocked daemon projections, scalar tables and duplicate terminal examples without equivalent replacement coverage. Per-source reviews record retained terminal content preservation, process ownership refusal and recovery boundaries together with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1516 removes package-electron-runtime-contract.test.mjs, a copied release-workflow platform and marker matrix. Its workflow wiring alarm is intentionally retired; no actual packaged watcher delivery was exercised and no equivalent replacement is claimed. Unaffected dated execution receipts retain their original scope; changed optional selectors have no fresh full-gate execution result. This audit removes the paired-web stopped-during-outage and late-ready direct-branch examples. The retained pre-ready cancellation, failed-unwatch retry and replay-continuity cases are not equivalent replacements for those ordering alarms. Current web-runtime-client.test.ts assertions below name only its retained pre-ready callback case; replay cases remain in their separate suite. Historical commands and results remain historical. Deterministic tests cover one shared healthy child per runtime process plus at most four bounded fault-quarantine children, native desktop, paired-runtime, and WSL snapshot processes sharing the same eight-physical-child reservation with typed serialized event-driven capacity recovery, including recursive re-wait when crash recovery reclaims an announced slot, one quarantine attempt per watch lifetime, generation-scoped cancellation, bounded termination with removable deadline waiters, replacement-crawl errors that cannot be blessed by a late readiness ack, pre-ready paired-web cancellation registered before an unbounded capacity wait, already-resolved and late physical-exit retry of a rejected paired-web teardown, physical desktop-install and local/SSH PTY teardown, stale-generation rejection, per-child stat bounds, final-overflow RPC delivery, and renderer eviction before terminal callbacks can retry. Destructive local and SSH removal fences both ID-derived and resolved-cwd terminal roots, including sibling-root/cwd combinations with reverse admission rollback, closes descendant relay watches before parent deletion, enumerates authoritative provider/cwd ownership, falls back to daemon spawn cwd before OSC 7, and keeps daemon/relay immediate shutdown pending until native exit or a bounded fail-closed error; a dead relay PID discovered during attach also settles concurrent shutdown before its stale entry is reaped, while Windows runtime watcher deadlines retain one removable physical-close waiter, treat Node's error-close path as positive physical-exit proof, and clear root ownership on late close. The relay policy test maps a standard repository's base, Git common directory, and worktree roots to one healthy supervisor, then proves a shared-child failure recovers those roots in separate quarantine supervisors with overflow and resumed events. Abortable capacity, quarantine, runtime-root, relay, SSH, and relay pre-install setup waits attach one reaction to each shared promise and explicitly remove 10,000 cancelled caller closures while one anchor remains. Built-entry macOS harnesses kill the desktop/runtime and SSH relay watcher children, then require automatic resubscription and later events while the host process and relay PTY survive. SSH bundles the child boundary beside relay.js, requires both artifacts for install completeness, and preserves registration-owned same-root cancellation. The test-value audit corrects the active renderer watch command/file reference to the existing runtime-file-client-watch.test.ts, whose three cases are retained unchanged. They observe synthetic subscription retry/fanout and unsubscribe/unwatch calls, not actual preload/main bookkeeping or physical watcher-process cleanup. The separate read/preview client suite loses forwarding, supplied-error and negative-only examples, retaining binary detection and Unicode SVG decoding. Historical evidence commands/results are unchanged. Test-value audit1293 removes src/main/providers/local-pty/local-pty-provider-session-inventory.test.ts (2 authored cases). Their isolated assertions no longer run; surviving tests are not claimed as equivalent coverage. Fresh execution against the candidate SHA is required. Audit1312 removes low-value imported specs from aggregate test drivers; old whole-driver execution receipts are archived. Retained registrations are unchanged. No equivalent replacement coverage is claimed. Audit1383: src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts reduced. Lose command-list/receipt/default/background activation matrices whose provider spawn returns canned IDs. Retained test proves actual tab allocation separation, not terminal startup. No equivalent replacement coverage is claimed. Existing exact unaffected receipts and their platform/provider/budget claims remain scoped to unchanged selectors. Changed optional selectors have no new execution result. Audit1401: src/main/runtime/orca-runtime-tests/mobile-session-tabs-duplicate-pty-repair.spec.ts removed. Loses listing-on-already-repaired-data alarm; it never established runtime repair invocation. src/main/runtime/orca-runtime-tests/mobile-session-tabs-split-agent-tab.spec.ts removed. Loses actual mobile split sibling shell versus recorded-agent resume command alarms; no equivalent coverage asserted. src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts removed. Loses separate setup/startup tab result in that manually assembled path. src/main/runtime/orca-runtime.test.ts reduced. Driver registration/import of a whole-deleted selected spec is removed; directly authored cases remain unchanged. No extra authored removal credit. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result. Audit1412: src/main/workspaces/removal/watcher-removal-gate.test.ts reduced. Loses exact removal-error compatibility with renderer predicate under shared message constants. Kept public refusal outcomes do not establish that renderer classification. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result.", + "coverageNotes": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2056 (sources2054/2055): Intentionally retire these low-value alarms without replacement: src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts: delivers remote file changes to listeners for the same worktree. src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/mobile-session-tabs-part-14.spec'). src/main/runtime/orca-runtime-tests/terminal-side-effect-facts-part-03.spec.ts: arms the stale-title timer for a seeded working title; arms the Command Code scrape from the noted spawn command; prefers the tracked title over a stale renderer lastTitle in the hydration seed; writes recent output into the current runtime output map; uses a later title handler replacement with the runtime receiver; calls the current injected listener with the runtime receiver; keeps optional reconciliation lazy after its callback is removed. src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-14.spec.ts: lists them, so a paired client can mirror a migrated editor; closes one by retiring it from the host session; keeps an unsaved draft unless the close is forced; shares aggregate work but leaves a replacement pending promise alone; lets a targeted refresh finish while an aggregate refresh waits; refuses a close when the store lacks %s before the guard; keeps a later missing %s on the original caught failure path. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Exact retired assertionRefs, preserved as historical objects only: {\n \"file\": \"src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts\",\n \"assertions\": [\n \"terminal watcher error and setup failure evict the shared entry before retrying callbacks run\",\n \"an old stream end cannot evict the replacement stream, which still delivers to its listener\",\n \"two same-worktree listeners share a synthetic subscription until the final stop unsubscribes and sends unwatch\"\n ]\n } Historical prior account (preserved verbatim): Audit1973: Delete the standalone mocked watcher cleanup suite and its entire sole-file prospective command. Retire the removed file-command transport-abort alias; preserve the independent cancellation suite and final-overflow/error/end cleanup callback, and13 unrelated receipt objects. Whole original changed dated receipts are retired, never narrowed or transferred. This metadata review claims no fresh execution, equivalent replacement, provider/PTY, live remote/mobile, rendered, cross-platform, soak or measured-budget result. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Partial runtime files-watch suite retains controlled Windows close-event waiting, shared-client survival, late physical-exit ownership/deadline cleanup and shutdown joining. Its combined Windows close/deadline/late-exit alias remains backed by unchanged callbacks. Retire native-error-close, pre-publication setup ownership and failed-delete rearm aliases; associated delegation, failure forwarding, removal routing and recovery examples are lost. These tests inject watch/subscription seams and do not observe an actual native watcher or physical process. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1925 (sources1922/1923): Partial runtime watcher-host suite loses mocked option/event/overflow/resubscription forwarding, supplied teardown-error propagation, aborted-signal/no-call and private ten-thousand-waiter-count inventories. Retire the 10,000-caller bound and failed-recovery-subscriber-ending assertion aliases. Keep same-root owner cancellation, pre-ready replacement, held root-close join, stale-generation recovery delivery and initial failed-root ownership mappings backed by retained real-host callbacks with injected subscription/error/physical-exit promises. No actual native watcher, process termination or resource bound was observed. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1683 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1645 removes low-value test projections, mock wiring, private-state assertions, inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained public behavior and refusal boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1640 removes low-value IPC, SSH, workspace and transcript-watcher tests without equivalent replacement coverage. Per-source reviews record retained public refusal, cancellation, execution-host and real-filesystem outcomes with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1634 removes low-value IPC, filesystem and runtime projections, argument inventories and duplicate examples without equivalent replacement coverage. Per-source reviews record retained cancellation, ownership, authorization, filesystem and SQLite boundaries with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1627 removes low-value mocked daemon projections, scalar tables and duplicate terminal examples without equivalent replacement coverage. Per-source reviews record retained terminal content preservation, process ownership refusal and recovery boundaries together with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1623 removes low-value mocked daemon projections, scalar tables and duplicate terminal examples without equivalent replacement coverage. Per-source reviews record retained terminal content preservation, process ownership refusal and recovery boundaries together with injected/native/remote limits. Affected historical receipts are retired; gates without active current evidence are downgraded. Historical scope before this audit: Audit1516 removes package-electron-runtime-contract.test.mjs, a copied release-workflow platform and marker matrix. Its workflow wiring alarm is intentionally retired; no actual packaged watcher delivery was exercised and no equivalent replacement is claimed. Unaffected dated execution receipts retain their original scope; changed optional selectors have no fresh full-gate execution result. This audit removes the paired-web stopped-during-outage and late-ready direct-branch examples. The retained pre-ready cancellation, failed-unwatch retry and replay-continuity cases are not equivalent replacements for those ordering alarms. Current web-runtime-client.test.ts assertions below name only its retained pre-ready callback case; replay cases remain in their separate suite. Historical commands and results remain historical. Deterministic tests cover one shared healthy child per runtime process plus at most four bounded fault-quarantine children, native desktop, paired-runtime, and WSL snapshot processes sharing the same eight-physical-child reservation with typed serialized event-driven capacity recovery, including recursive re-wait when crash recovery reclaims an announced slot, one quarantine attempt per watch lifetime, generation-scoped cancellation, bounded termination with removable deadline waiters, replacement-crawl errors that cannot be blessed by a late readiness ack, pre-ready paired-web cancellation registered before an unbounded capacity wait, already-resolved and late physical-exit retry of a rejected paired-web teardown, physical desktop-install and local/SSH PTY teardown, stale-generation rejection, per-child stat bounds, final-overflow RPC delivery, and renderer eviction before terminal callbacks can retry. Destructive local and SSH removal fences both ID-derived and resolved-cwd terminal roots, including sibling-root/cwd combinations with reverse admission rollback, closes descendant relay watches before parent deletion, enumerates authoritative provider/cwd ownership, falls back to daemon spawn cwd before OSC 7, and keeps daemon/relay immediate shutdown pending until native exit or a bounded fail-closed error; a dead relay PID discovered during attach also settles concurrent shutdown before its stale entry is reaped, while Windows runtime watcher deadlines retain one removable physical-close waiter, treat Node's error-close path as positive physical-exit proof, and clear root ownership on late close. The relay policy test maps a standard repository's base, Git common directory, and worktree roots to one healthy supervisor, then proves a shared-child failure recovers those roots in separate quarantine supervisors with overflow and resumed events. Abortable capacity, quarantine, runtime-root, relay, SSH, and relay pre-install setup waits attach one reaction to each shared promise and explicitly remove 10,000 cancelled caller closures while one anchor remains. Built-entry macOS harnesses kill the desktop/runtime and SSH relay watcher children, then require automatic resubscription and later events while the host process and relay PTY survive. SSH bundles the child boundary beside relay.js, requires both artifacts for install completeness, and preserves registration-owned same-root cancellation. The test-value audit corrects the active renderer watch command/file reference to the existing runtime-file-client-watch.test.ts, whose three cases are retained unchanged. They observe synthetic subscription retry/fanout and unsubscribe/unwatch calls, not actual preload/main bookkeeping or physical watcher-process cleanup. The separate read/preview client suite loses forwarding, supplied-error and negative-only examples, retaining binary detection and Unicode SVG decoding. Historical evidence commands/results are unchanged. Test-value audit1293 removes src/main/providers/local-pty/local-pty-provider-session-inventory.test.ts (2 authored cases). Their isolated assertions no longer run; surviving tests are not claimed as equivalent coverage. Fresh execution against the candidate SHA is required. Audit1312 removes low-value imported specs from aggregate test drivers; old whole-driver execution receipts are archived. Retained registrations are unchanged. No equivalent replacement coverage is claimed. Audit1383: src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts reduced. Lose command-list/receipt/default/background activation matrices whose provider spawn returns canned IDs. Retained test proves actual tab allocation separation, not terminal startup. No equivalent replacement coverage is claimed. Existing exact unaffected receipts and their platform/provider/budget claims remain scoped to unchanged selectors. Changed optional selectors have no new execution result. Audit1401: src/main/runtime/orca-runtime-tests/mobile-session-tabs-duplicate-pty-repair.spec.ts removed. Loses listing-on-already-repaired-data alarm; it never established runtime repair invocation. src/main/runtime/orca-runtime-tests/mobile-session-tabs-split-agent-tab.spec.ts removed. Loses actual mobile split sibling shell versus recorded-agent resume command alarms; no equivalent coverage asserted. src/main/runtime/orca-runtime-tests/worktree-native-chat-setup.spec.ts removed. Loses separate setup/startup tab result in that manually assembled path. src/main/runtime/orca-runtime.test.ts reduced. Driver registration/import of a whole-deleted selected spec is removed; directly authored cases remain unchanged. No extra authored removal credit. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result. Audit1412: src/main/workspaces/removal/watcher-removal-gate.test.ts reduced. Loses exact removal-error compatibility with renderer predicate under shared message constants. Kept public refusal outcomes do not establish that renderer classification. No equivalent replacement coverage is claimed. Existing exact unaffected receipts remain unchanged and apply only to their unchanged selectors. Changed optional selectors have no new execution result.", "motivatingLinks": [ "https://github.com/stablyai/orca/issues/5308", "https://github.com/stablyai/orca/issues/7547", "https://github.com/stablyai/orca/issues/8212" ], "invariant": "Local runtime and SSH relay roots share one watcher child per host process while healthy and use at most four quarantine children after shard failure; native desktop, paired-runtime, and direct WSL snapshot children consume the same process-wide eight-child reservation and retry transient exhaustion without permanently disabling a root. Cancellation remains owned by its child generation and settles only after acknowledgement or bounded physical exit; paired-web retains cleanup ownership and can cancel before readiness even while capacity is exhausted, retries a failed unwatch after already-resolved or late physical exit without duplicating stream termination, replacements never overlap a still-retiring generation, and aborted or timed-out callers detach immediately even while queued behind same-root relay teardown. Destructive local and SSH cleanup holds one containment-aware watcher/terminal removal fence through descendant watcher and PTY physical exit plus parent Git/filesystem deletion; daemon and relay kill requests retain ownership until the actual subprocess exit callback or an equivalent positive dead-PID proof and propagate rejected signals or a bounded failure otherwise. A crash-fused shared shard moves affected roots into the bounded quarantine pool, a second fused quarantine ends those watch lifetimes, interrupted setup emits overflow after recovery, metadata work stays bounded, and the SSH relay, PTYs, Git, and other filesystem operations never load the watcher native addon or share its fault boundary.", - "oracle": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2056 (sources2054/2055): Intentionally retire these low-value alarms without replacement: src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts: delivers remote file changes to listeners for the same worktree. src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/mobile-session-tabs-part-14.spec'). src/main/runtime/orca-runtime-tests/terminal-side-effect-facts-part-03.spec.ts: arms the stale-title timer for a seeded working title; arms the Command Code scrape from the noted spawn command; prefers the tracked title over a stale renderer lastTitle in the hydration seed; writes recent output into the current runtime output map; uses a later title handler replacement with the runtime receiver; calls the current injected listener with the runtime receiver; keeps optional reconciliation lazy after its callback is removed. src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-14.spec.ts: lists them, so a paired client can mirror a migrated editor; closes one by retiring it from the host session; keeps an unsaved draft unless the close is forced; shares aggregate work but leaves a replacement pending promise alone; lets a targeted refresh finish while an aggregate refresh waits; refuses a close when the store lacks %s before the guard; keeps a later missing %s on the original caught failure path. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Exact retired assertionRefs, preserved as historical objects only: {\n \"file\": \"src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts\",\n \"assertions\": [\n \"terminal watcher error and setup failure evict the shared entry before retrying callbacks run\",\n \"an old stream end cannot evict the replacement stream, which still delivers to its listener\",\n \"two same-worktree listeners share a synthetic subscription until the final stop unsubscribes and sends unwatch\"\n ]\n } Historical prior account (preserved verbatim): Audit1973: Delete the standalone mocked watcher cleanup suite and its entire sole-file prospective command. Retire the removed file-command transport-abort alias; preserve the independent cancellation suite and final-overflow/error/end cleanup callback, and13 unrelated receipt objects. Whole original changed dated receipts are retired, never narrowed or transferred. This metadata review claims no fresh execution, equivalent replacement, provider/PTY, live remote/mobile, rendered, cross-platform, soak or measured-budget result. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Partial runtime files-watch suite retains controlled Windows close-event waiting, shared-client survival, late physical-exit ownership/deadline cleanup and shutdown joining. Its combined Windows close/deadline/late-exit alias remains backed by unchanged callbacks. Retire native-error-close, pre-publication setup ownership and failed-delete rearm aliases; associated delegation, failure forwarding, removal routing and recovery examples are lost. These tests inject watch/subscription seams and do not observe an actual native watcher or physical process. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1925 (sources1922/1923): Partial runtime watcher-host suite loses mocked option/event/overflow/resubscription forwarding, supplied teardown-error propagation, aborted-signal/no-call and private ten-thousand-waiter-count inventories. Retire the 10,000-caller bound and failed-recovery-subscriber-ending assertion aliases. Keep same-root owner cancellation, pre-ready replacement, held root-close join, stale-generation recovery delivery and initial failed-root ownership mappings backed by retained real-host callbacks with injected subscription/error/physical-exit promises. No actual native watcher, process termination or resource bound was observed. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1683 retires changed-source assertion and execution scope without equivalent replacement coverage. Existing unchanged-selector receipts remain scoped to their original sources. Historical prior-source oracle: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: Fill the global watcher-child reservation, require native desktop, paired-runtime, and WSL setup to retain typed transient capacity and retry after one serialized release, then cancel pending desktop and paired-client watches and require no retry. Hold one live anchor while aborting 10,000 callers at the capacity, quarantine-assignment, runtime-root, relay-setup, and SSH-registration layers; require exactly one explicit waiter and one base-promise reaction to remain, then zero after settlement. Hold an old relay unsubscribe, queue one same-root setup leader and one anchor, then abort 10,000 joiners; require every cancellation to settle before teardown releases, exactly one anchor waiter, and only one later physical install. Advance the former paired-web callback deadline while capacity remains full, require cancellation ownership to stay live, publish the server token before ready, and require files.unwatch to abort setup before a later slot release can install a native root. Reject the first ready-watch unwatch with a physical-exit promise, resolve that exit later, and require a fresh physical unwatch attempt, registry release, and exactly one stream end. Fill one healthy shared child with multiple roots, map a standard SSH repository's base, Git common directory, and worktree roots to one relay supervisor, then inject IPC disconnect-before-exit, queued and active cancellation, operations during termination, kill-without-exit, repeated unkillable generations, initial and recovery timeouts, and shard failure. During local and SSH deletion, attempt a late terminal spawn whose worktree identity and cwd name different roots, enumerate a non-prefix remote PTY by authoritative worktree ownership, withhold daemon/relay exit callbacks, make attach positively prove a concurrent-shutdown PID dead, and make the provider unavailable; require every acquired admission to roll back on rejection, dead-PID proof to settle physical exit before reaping, timeout to retain ownership, and Git/filesystem deletion never to start. Require bounded watcher process count, physical-exit settlement, healthy-root recovery, stale-generation rejection, overflow plus later events, and fail-closed destructive cleanup. Verify relay build hashes and remote install probes require both relay.js and relay-watcher.js. Launch the built relay with a real PTY and Parcel watch, kill only the watcher child, and require the same relay PID and PTY to answer before a replacement child delivers overflow and a post-crash event.", + "oracle": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The retained runtime registration driver loses three complete imported fragments: changed-working-label title facts and suppressed-frame tracked snapshots; refusing SSH GitLab MR-base lookup before local Git/glab calls; and missing folder-workspace path creation/update refusal before spawn/store mutation. These alarms are intentionally lost. Other imported fragments and separately rooted dated receipts are unchanged; they are not equivalent replacement coverage. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2090 (sources2088/2089): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2086 (sources2082/2083): Current reduced suites intentionally retire the alarms listed in knownGaps; prior dated execution and replacement claims do not validate changed source bytes or selectors. Historical prior account (preserved verbatim): Audit2076 (sources2073/2074): Intentionally retire these low-value alarms without replacement: src/main/runtime/orca-runtime-tests/terminal-handles-part-02.spec.ts: recovers a moved SSH pane through the tab it now sits in, not the one its lease froze; reports a plain expired orphan; does not report a pane whose only recent lease was superseded; does not report a pane whose only recent lease had its relay id recycled; picks the eligible orphan over a superseded predecessor that matches first; collects the same tabs the per-tab reader reports, in one sweep of the leases; reports no candidate when no lease names any of the worktree tabs. src/main/runtime/orca-runtime-tests/runtime-availability.spec.ts: starts unavailable with no authoritative window; reports the configured Windows terminal shell on status; reports floating workspace availability from settings on status; polls floating tabs with targeted PTY liveness and no repo/provider inventory; hydrates persisted tabs when the store cannot report repos; advertises browser screencast only when a renderer window is available; routes mobile Codex reset consumption through the account mutation coordinator; maps a definite pre-provider rejection into an authoritative current snapshot. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Frozen current literal-import graph reaches these changed test bytes; unresolved local and computed edges remain debt, and current reachability is not historical execution or equivalent replacement evidence. Historical prior account (preserved verbatim): Audit2056 (sources2054/2055): Intentionally retire these low-value alarms without replacement: src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts: delivers remote file changes to listeners for the same worktree. src/main/runtime/orca-runtime.test.ts: ordinary registration cleanup removes only await import('./orca-runtime-tests/mobile-session-tabs-part-14.spec'). src/main/runtime/orca-runtime-tests/terminal-side-effect-facts-part-03.spec.ts: arms the stale-title timer for a seeded working title; arms the Command Code scrape from the noted spawn command; prefers the tracked title over a stale renderer lastTitle in the hydration seed; writes recent output into the current runtime output map; uses a later title handler replacement with the runtime receiver; calls the current injected listener with the runtime receiver; keeps optional reconciliation lazy after its callback is removed. src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-14.spec.ts: lists them, so a paired client can mirror a migrated editor; closes one by retiring it from the host session; keeps an unsaved draft unless the close is forced; shares aggregate work but leaves a replacement pending promise alone; lets a targeted refresh finish while an aggregate refresh waits; refuses a close when the store lacks %s before the guard; keeps a later missing %s on the original caught failure path. Existing unrelated evidenceRuns retain their original scope; unchanged dated receipt objects do not validate changed source bytes or reduced selectors. No fresh gate execution or historical import-closure equivalence is claimed. Exact retired assertionRefs, preserved as historical objects only: {\n \"file\": \"src/renderer/src/runtime/files/watching/runtime-file-client-watch.test.ts\",\n \"assertions\": [\n \"terminal watcher error and setup failure evict the shared entry before retrying callbacks run\",\n \"an old stream end cannot evict the replacement stream, which still delivers to its listener\",\n \"two same-worktree listeners share a synthetic subscription until the final stop unsubscribes and sends unwatch\"\n ]\n } Historical prior account (preserved verbatim): Audit1973: Delete the standalone mocked watcher cleanup suite and its entire sole-file prospective command. Retire the removed file-command transport-abort alias; preserve the independent cancellation suite and final-overflow/error/end cleanup callback, and13 unrelated receipt objects. Whole original changed dated receipts are retired, never narrowed or transferred. This metadata review claims no fresh execution, equivalent replacement, provider/PTY, live remote/mobile, rendered, cross-platform, soak or measured-budget result. Historical prior account (preserved verbatim): Audit1931 (sources1928/1929): Partial runtime files-watch suite retains controlled Windows close-event waiting, shared-client survival, late physical-exit ownership/deadline cleanup and shutdown joining. Its combined Windows close/deadline/late-exit alias remains backed by unchanged callbacks. Retire native-error-close, pre-publication setup ownership and failed-delete rearm aliases; associated delegation, failure forwarding, removal routing and recovery examples are lost. These tests inject watch/subscription seams and do not observe an actual native watcher or physical process. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1925 (sources1922/1923): Partial runtime watcher-host suite loses mocked option/event/overflow/resubscription forwarding, supplied teardown-error propagation, aborted-signal/no-call and private ten-thousand-waiter-count inventories. Retire the 10,000-caller bound and failed-recovery-subscriber-ending assertion aliases. Keep same-root owner cancellation, pre-ready replacement, held root-close join, stale-generation recovery delivery and initial failed-root ownership mappings backed by retained real-host callbacks with injected subscription/error/physical-exit promises. No actual native watcher, process termination or resource bound was observed. Preserve meaningful remaining selectors/assertions and unchanged dated receipts within their original scope. Reduced selectors have no fresh gate execution evidence in this metadata packet. No equivalent replacement, physical watcher/PTY/provider, live SSH/remote/mobile, rendered, cross-platform, soak or measured-budget result is claimed. Historical prior account (preserved verbatim): Audit1683 retires changed-source assertion and execution scope without equivalent replacement coverage. Existing unchanged-selector receipts remain scoped to their original sources. Historical prior-source oracle: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: Existing exact unchanged-selector receipts remain the only current execution evidence. Changed source selectors have no fresh execution result. Historical prior-source oracle account: Fill the global watcher-child reservation, require native desktop, paired-runtime, and WSL setup to retain typed transient capacity and retry after one serialized release, then cancel pending desktop and paired-client watches and require no retry. Hold one live anchor while aborting 10,000 callers at the capacity, quarantine-assignment, runtime-root, relay-setup, and SSH-registration layers; require exactly one explicit waiter and one base-promise reaction to remain, then zero after settlement. Hold an old relay unsubscribe, queue one same-root setup leader and one anchor, then abort 10,000 joiners; require every cancellation to settle before teardown releases, exactly one anchor waiter, and only one later physical install. Advance the former paired-web callback deadline while capacity remains full, require cancellation ownership to stay live, publish the server token before ready, and require files.unwatch to abort setup before a later slot release can install a native root. Reject the first ready-watch unwatch with a physical-exit promise, resolve that exit later, and require a fresh physical unwatch attempt, registry release, and exactly one stream end. Fill one healthy shared child with multiple roots, map a standard SSH repository's base, Git common directory, and worktree roots to one relay supervisor, then inject IPC disconnect-before-exit, queued and active cancellation, operations during termination, kill-without-exit, repeated unkillable generations, initial and recovery timeouts, and shard failure. During local and SSH deletion, attempt a late terminal spawn whose worktree identity and cwd name different roots, enumerate a non-prefix remote PTY by authoritative worktree ownership, withhold daemon/relay exit callbacks, make attach positively prove a concurrent-shutdown PID dead, and make the provider unavailable; require every acquired admission to roll back on rejection, dead-PID proof to settle physical exit before reaping, timeout to retain ownership, and Git/filesystem deletion never to start. Require bounded watcher process count, physical-exit settlement, healthy-root recovery, stale-generation rejection, overflow plus later events, and fail-closed destructive cleanup. Verify relay build hashes and remote install probes require both relay.js and relay-watcher.js. Launch the built relay with a real PTY and Parcel watch, kill only the watcher child, and require the same relay PID and PTY to answer before a replacement child delivers overflow and a post-crash event.", "commands": [ "pnpm exec vitest run --config config/vitest.config.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-unsubscribe-timeout.test.ts --reporter=dot", "pnpm exec vitest run --config config/vitest.config.ts src/main/ipc/filesystem/watchers/filesystem-watcher.test.ts src/main/ipc/filesystem/watchers/filesystem-watcher-native-capacity.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-child-registry.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-child-launch.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-process.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-supervisor-capacity.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-disconnect-termination.test.ts src/main/ipc/filesystem/watcher-process/runtime-watcher-pending-assignment.test.ts src/main/ipc/filesystem/watcher-process/parcel-watcher-event-delivery.test.ts src/main/ipc/filesystem/watchers/filesystem-watcher-local-unsubscribe.test.ts src/main/ipc/filesystem/watchers/filesystem-watcher-wsl.test.ts src/main/ipc/worktrees-delete-pty-teardown.test.ts src/main/ipc/worktrees-orphan-directory-cleanup.test.ts src/main/ipc/worktrees-remove-preflight.test.ts src/main/workspaces/removal/watcher-removal-gate.test.ts src/main/runtime/files/watch/orca-runtime-files-watch.test.ts src/main/runtime/workspaces/removal/worktree-teardown.test.ts src/main/runtime/construction/orca-runtime.test.ts src/main/runtime/rpc/methods/files/files.test.ts src/main/runtime/rpc/methods/files/files-watch-cancellation.test.ts src/renderer/src/web/web-runtime-client.test.ts src/shared/cross-platform-path.test.ts src/shared/physical-exit-tracker.test.ts src/shared/promise-settlement-waiters.test.ts src/main/ipc/pty/pty-spawn-cwd-fallback.test.ts src/main/providers/local-pty/local-pty-provider-shutdown.test.ts src/main/daemon/providers/adapter/testing/daemon-pty-adapter.test.ts src/main/daemon/pty-subprocess/testing/pty-subprocess.test.ts src/main/daemon/sessions/testing/session.test.ts src/main/daemon/terminal-host/testing/terminal-host.test.ts src/main/daemon/terminal-host/testing/terminal-host-session-reaping-leak.test.ts --reporter=dot", @@ -12820,7 +12825,9 @@ "Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. All prior notes, assertion references and complete dated receipts retain their original historical scopes.", "Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. All prior notes and archived receipts retain original scopes.", "Audit3460 final-v2 composes accepted AUTHOR3454/PEER3455 on actual post263 main: intentionally removes99 authored/106 potential expanded cases across16 selected drivers, including7 whole suites, retaining24 authored/26 potential cases. No equivalent replacement or fresh runtime is claimed. Broad root-unit discovery loses whole cross-client workspace-sync, terminal-preview, picker-skill-hint and close-routing suites plus reduced worker-stop, remote-owned-record, scanner, mobile-bundle and daemon-health samples. Separate E2E discovery also loses the paired mobile/desktop mirror journey and two unit drivers. Retained callback/table/outer-loop/guard bytes are exact; retained checks do not replace the removed alarms. No app, native device, mixed-version, remote or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", - "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." + "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", + "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. All prior notes and dated/archived evidence retain original scopes.", + "Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." ], "demotionRule": "Demote or quarantine if the fault harness flakes without a product or harness bug, if healthy operation exceeds one runtime watcher child, if total physical operation exceeds eight children including retiring generations, if quarantine children outlive their roots or repeat after fusing, if event delivery becomes unbounded, or if metadata/stat work returns to the serve process." }, @@ -14058,10 +14065,10 @@ "providers": ["paired-runtime"], "coveredPlatforms": ["linux"], "coveredProviders": ["paired-runtime"], - "coverageNotes": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit1579 removes workflow/manifest/regex/private binary-header and diagnostic metric inventories, fixture selectors, injected forwarding/platform/signature/process examples, unproven overlap and internal churn counters without equivalent replacement coverage. Retains actual Git trust/diffs; filesystem hash/ELF-region/native discovery; child CLI module repair; compiler/lock/publication/rollback and synthetic Electron install/cache ownership; executable shell quoting/exit; controlled notary/timer cleanup; local patched extractor file/AST freshness; real Chromium/WebKit CSP/sandbox/HTTPS image/referrer/link/preservation controls under parsed Android headers. Five unchanged setup-pnpm/VM request-cleanup definitions reviewed and not rerun. GitHub, installer/archive, OS metric, client, native delegate, compressor, builder and Apple boundaries injected or absent where recorded. No actual native PTY, shipped full/native app, real Actions/cache/npm signature authority, downloaded Electron archive, real objdump/linker/native CLI/AppImage, signing/NSIS/notary/upload/updater or live freeze proof. Historical scope before this audit: Published Linux 1.4.188 desktop against current source in both directions; scheduled weekly and manually runnable. No required PR check.", + "coverageNotes": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit1579 removes workflow/manifest/regex/private binary-header and diagnostic metric inventories, fixture selectors, injected forwarding/platform/signature/process examples, unproven overlap and internal churn counters without equivalent replacement coverage. Retains actual Git trust/diffs; filesystem hash/ELF-region/native discovery; child CLI module repair; compiler/lock/publication/rollback and synthetic Electron install/cache ownership; executable shell quoting/exit; controlled notary/timer cleanup; local patched extractor file/AST freshness; real Chromium/WebKit CSP/sandbox/HTTPS image/referrer/link/preservation controls under parsed Android headers. Five unchanged setup-pnpm/VM request-cleanup definitions reviewed and not rerun. GitHub, installer/archive, OS metric, client, native delegate, compressor, builder and Apple boundaries injected or absent where recorded. No actual native PTY, shipped full/native app, real Actions/cache/npm signature authority, downloaded Electron archive, real objdump/linker/native CLI/AppImage, signing/NSIS/notary/upload/updater or live freeze proof. Historical scope before this audit: Published Linux 1.4.188 desktop against current source in both directions; scheduled weekly and manually runnable. No required PR check.", "motivatingLinks": ["https://github.com/stablyai/orca/actions/runs/34069063016"], "invariant": "A paired client and host without client-hosted browser capabilities retain server-hosted browser placement across supported version skew.", - "oracle": "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Require both existing named browser placement scenarios to pass three times with one attempt, zero skips, zero failures, and no report errors.", + "oracle": "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. Audit3437 composes ROOT-adopted3431/independent3432: intentionally removes127 authored/174 potential expanded cases across12 sources, deleting7 whole suites and one267line unused helper without test credit. Retain8 authored/13 potential cases. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses reduced or deleted relay, coordinator mail, recovery watchdog, automation observer, launch picker, folder PTY, profile restart, web migration, retirement, pointer delivery, agent-row and run-bound mailbox examples. Retained checks cover relay hello/payload and SIGTERM drain, coordinator pending-check and clear-to-Run continuity, profile persistence, the full six-row web merge table and real run-bound delivery waiter. These do not replace removed alarms. Audit3429 composes root-adopted3426/independent3427: intentionally removes168 authored/205 potential expanded cases across16 semantic sources, deleting11 whole suites and one304line unused assertion helper. Five authored/five potential cases remain; the duplicate500k graph row is dropped while its50k row and callback remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted migration/recovery/federation/rich-draft/paste/automation/mobile/catalog examples. Keep server-edit preservation during migration, native supervised cancellation holding the agent home until physical stop, pagehide lost-text preservation, graph old-result WRAPPER WeakRef collection with50k rows, and mobile React render isolation. The WeakRef does not prove all commit objects are collectible or a memory budget. Audit3425 composes root-adopted3420/independent3421: intentionally removes83 authored/112 potential expanded registrations across13 semantic sources, deleting7 whole conventional suites and1 registering helper;7 authored/10 potential expanded cases remain. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses the adopted mock orchestration, refusal, restart and storage examples. Retain unchanged two-phase real-WSL deployment/survival, the reduced idle-release holder proof without restart/first-output measurement, real worker stop/release cleanup barrier rows, physical Git branch-preservation rows, and SSH disconnected-close proof. Other SSH and CLI variants and WSL telemetry/warmup churn are intentionally removed. Audit3417 composes root-adopted3415v2/independent3416v2: intentionally removes117authored/173potential expanded registrations from12semantic sources, deleting4whole conventional suites and3registration modules and retaining4authored/5expanded cases, without equivalent replacement or fresh execution. Broad root-unit discovery loses naming/title/queued-host fixtures, main-window focus-shortcut and attach-services mock orchestration, Git Bash console-capacity fake process inventories, most real-Git preparation/locking variants and Claude/Codex catch-up examples. Keep lifetime rejected-writer followed by late termination proof and autonomous real-checkout cleanup; two-row real-Git rollback after a real move preserving a replacement/admin marker or revoked permission; a live translator write during a held transcript read fencing stale Claude history append; and Codex expected-item revision rejection before a historical writer shim. Codex driver registers retained bounds exactly once with original inputs; bounds does not prove modern writer persistence. Accepted changed-lock preflight sentinel and omitted-parent routing losses are not replaced. Audit3413 composes root-adopted3405/independent3406 and3409/independent3410: intentionally removes337 authored registrations/363potential expanded cases from13sources, retaining31authored/32potential cases and deleting one whole clipboard suite, without equivalent replacement or fresh execution. Broad root-unit discovery loses35WSL examples,22STT examples,22shared-server-banner examples,18clipboard-handler examples (whole suite deleted),14VM lifecycle examples and16commit-message cancellation examples. Retain4WSL and7STT mocked connector/worker cases,1rendered stale-banner refusal,2VM guard/failed-compensation cases and1late-stream ownership cancellation case; real WSL/microphone/remote-provider proof is not established. Separately discovered e2e-unit drivers lose13cross-version-wire local cases and15mobile active-publication cases, keeping same-process disk-reopen/new-host cursor content/epoch/sequence outcomes and a host/renderer-snapshot selection and inventory revision race respectively. Imported released stop-note scenarios remain unchanged and receive no duplicate credit. Cloud suites are outside root UNIT_INCLUDE. Audit3403 composes root-adopted3393/3394,3400/3401 and3398/3399: intentionally retires167 authored registrations/196 potential expanded cases across93 whole suites without equivalent replacement or fresh execution. Retained65 registrations/79 potential cases include one conditional missing-input failure sentinel; potential counts are not executions. Historical notes and complete dated receipts retain their exact original scopes. The broad root-unit command loses the UNIT_INCLUDE src/ subset of retired examples; mobile is outside UNIT_INCLUDE. Other lost alarms are recorded by the adopted per-case source reviews. Audit3390 composes root-adopted3382/independent3383 and3386/independent3387 and3388/independent3389: intentionally retires216 authored registrations/230 expanded cases in42 whole authored suites plus5 zero-case registration drivers, without equivalent replacement. Lost local alarms include fake PTY identity/routing/startup/replay/mode/title/mouse/IME/scroll matrices, broad startup identity and file-open wiring, mock mailbox title recovery, mobile row parity and whole-v-selective save parity. Retained24 registrations/26 expanded cases in22 suites preserve actual xterm Windows buffer/wrap/scroll, typed TOML scalar, real Promise cancellation/settlement, HTTP Codex scanner/projector semantics, durable recovery admission/newer-schema bytes, federation completion/ACK/SQLite atomicity and failed stale-run-write versus imported durable history. Additional retained clock-domain scheduling, durable freeze/ACK/checkpoint ordering, ABA dirty intent, SQL atomicity, folder-domain retier/last-note deletion, scrollback durable no-sync staging, PanelBridge subscriber lifetime and reserved-send publication preserve narrower concrete behaviors. Current finite owner drift assessed in bounded defining bodies; retained callback bytes unchanged. Historical notes and complete dated receipts keep their original exact scopes. No new execution, equivalent replacement, native focus/device/pixel, live remote or global mixed-version guarantee is claimed. Audit3381 composes root-adopted3376/independent3377 and3379/independent3380: intentionally retires436 authored registrations/670 expanded cases, including79 whole suites and12877 test lines, without equivalent replacement. Lost alarms include scheduling, account/host/status/filter/format inventories, microphone/image cleanup, title/bell/capability lifecycle routing, access/impersonation/outcome matrices, paste/input resend/ledger/retention, stale completions/reload containment, trust-read failure and popup OAuth/security/geometry vectors. Retained30 registrations/34expanded cases in22 suites preserve narrower SQLite TEMP/TZ/WAL, HTTP transcript root resume, mobile original-index Choose/Submit, child environment/Git lock expiry, stale ACK/PTY lifetime/peer fairness/durable replay/output ordering, actual DOM paste/editor focus, DEC2026 conservation, filesystem rollback, owned buffer lifetime, actual OSC parser/tracker handback, POSIX Python UTF8/fallback, durable trust concurrency/version, OpenSSH verdicts and hidden Chromium POST/route lifecycle. Mocked boundaries limit claims. Current bounded defining-owner reads preserve the retained oracles; captured identities are not full transitive semantic review. Both component blind pins authenticated before exchange; final reconciliation and explicit postpin3380reason correction supplement control, without retroactive blind claims. Root unit command excludes mobile and separately admitted Electron probes; use existing respective runners. No fresh execution, native focus/device/pixel UI, live SSH/remote, equivalent replacement or broad mixed-version/cross-platform validation is claimed. Prior dated receipts retain exact original scopes. Audit3375 composes root-adopted3370/independent3371 and3373/independent3374: intentionally retires79 authored registrations/84 expanded cases, including39 whole suites, without equivalent replacement. Lost alarms include federation field/parser inventories; Claude rejected-key precedence, credential clone redaction, managed-versus-user hook preservation and CLI defaults; setting/status/activity projections, mocked SSH/notification/reveal/Linear refusal, fake-clock reconnect and noise; image-fold artifact visibility, paste eligibility, setup trust, scripted ACP error-to-journal publication and ligature cache tuple aliasing; plus supplied folder-runner uniqueness, mobile host-overlay/credential/update/preference races, status/markup/title/keyed-row projections, subscriber cleanup, cancellation/close/refusal/link-issue and mirror suppression alarms. Retained3373 merger contiguity, dual-clock overdue-resume exactly-once, rejected-paste IPC sink and annotation schema roundtrip remain bounded. Retained actual HTTP/OSC status composition, destructive sibling-filesystem isolation, external contrast invariant and fixture CSS pixel intervention remain narrower. Original reviewed87authored/92expanded become9/9; current OpenCode suite adds two preserved upstream cases and the synthetic-title suite gained one Cursor policy mock-call case, independently reviewed for removal in root3378; current90/95 become11/11. That Cursor suppression/ready-frame regression alarm is intentionally lost; it does not observe native title ownership or tab flicker. For3370 only the peer was independently blind, with author exposure explicit;3373/3374 both blind pins were verified. This broad root unit command selects the reduced root unit suite; UNIT_INCLUDE excludes mobile tests, which require the existing mobile runner; retained Playwright component-pixel spec uses its separate existing headless fixture runner and is not validated by the broad unit command. No fresh execution, actual agent/shell/provider, OS material/focus, historical import equivalence, mobile, mixed-version or cross-platform evidence is claimed. All dated receipts retain their original scope. Audit3368 composes root-adopted3363/3364 and3367/3369: intentionally retires181 authored registrations/196 expanded cases, including28 whole suites, without equivalent replacement. Lost alarms include hibernation attention/liveness/backoff/sequential teardown and inventory, host launch compatibility/legacy/Kimi/resource cleanup, terminal recovery parser/pause/quarantine/remount, window-visibility omission-fence lifetime, launch-option rendered Start, sleeping-resume saved recipe/WSL quoting/cancellation, plus removed project/window/input variants. Additional tiny-suite losses include sentence/label/field/default inventories, credential omission, Windows argument/environment mapping, URI decoding, diagnostics, provider classification and fake-clock callback/parser examples. Retained ten registrations/twelve expanded cases include two literal Unicode Mn/Cf zero-width range guards in the actual character-range algorithm and protect narrower delayed editor bytes/destructive ownership, mounted hidden collision repair, ten transport epochs, duplicate-command refusal, caller timeout typing, executor unknown-opener refusal and failed-chunk user bytes. This broad unit command selects the reduced suite; no fresh runtime, real provider/network/PTY, filesystem restart, native UI, historical import equivalence, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3362 composes root-adopted3358/independent3360: intentionally retires129 authored registrations/140 expanded cases, including four whole suites, without equivalent replacement. Lost alarms include supplied orphan-adoption retry/malformed inventory/cancellation; fake-terminal replay parser/wedge/disposal callbacks; all-host catalog metadata and Windows preferences; host keepalive/protocol/stale/settings failures; and removed exact-stop/suppression, source save lifetime, diff-note queue, rename/editor close, input flood/chunk refusal/limits, purge and creation examples. Retained18 registrations/21 expanded cases protect narrower ownership, generation, dirty-state, queue convergence, encoding and attachment-lease outcomes. No replacement or keeper quota claim. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY/xterm parser, disk restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3359 composes root-adopted source3355/independent3356: intentionally retires130 authored registrations/139 expanded cases, including eight whole suites, without equivalent replacement. Lost alarms include strict create naming/capability/legacy payloads; supplied transport host replacement, deadline reattach, create-handoff and split activation; agent-session fork/clipboard/platform flags; shutdown cleared-scrollback/unmounted bindings; simulator host-collision/selection/shutdown; and removed editor, catalog, mutation, launch and status examples. Retained12 adverse-ordering/state-consequence cases are narrower and do not replace all removed alarms. Two queue absences receive zero removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH/PTY, filesystem restart, native UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3349 composes adopted batches3343,3345 and refreshed3354: intentionally retires234 authored registrations/294 expanded cases, including13 whole suites, without equivalent replacement. Lost alarms include supplied terminal reconnect/replacement/typeahead/mirror/result and snapshot examples; synthetic render-desync persistence/recovery; worktree listing/owner/catalog/lineage/removal and activation/creation examples; host-scoped PR and task-record projections; editor diff/status/comment and UI send-target examples; automation observation/wake, supplied sleep/deadline clock and native-chat completion/notification/relaunch samples; held-tab layout and cleanup/progress guard examples. Retained adverse-ordering and wire-boundary cases are narrower and do not replace all removed alarms. The already-upstream visibility-resume deletion receives zero new removal credit. This broad unit command selects the reduced suite; no fresh execution, historical import equivalence, real provider/network/SSH, native process/power/GPU, rendered UI, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain their original scope. Audit3342 composes adopted source batches3333/3334,3336/3337 and3340/3341: intentionally retires180 authored registrations/263 expanded cases, including seven whole suites, without equivalent replacement. Lost alarms include corrupt editor hydration; mobile editor/diff publication and privacy projections; stale-resubscribe routing; model-picker projection/refetch/retry/fallback; editor check-details and frame property inventories; latched-pane cleanup/quiescence; client create replay identity/options/compatibility; and removed owner, async routing, terminal and layout examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts keep their original scope. Audit3335 composes adopted source unions3327/3328 and3330/3332: intentionally retires100 authored registrations/123 expanded cases, including ten whole suites, without equivalent replacement. Lost alarms include seeded host persistence/contention and stale-host cleanup; browser snapshot assembly; mocked wake/DPR callbacks and frame budgets; preflight projections; palette authority/cache suppression, URL intent and saved-title cross-repo collisions; metadata coverage/set patch and dropdown assembly; and removed terminal, worktree, review-refresh and owner-routing examples. Retained definitions do not replace these alarms. This broad root unit command selects the reduced suite; no fresh execution, historical source/import equivalence, real provider/network/SSH, rendered UI/GPU/focus, durability restart, mobile, mixed-version or cross-platform validation is claimed. Unrelated dated receipts retain only their original scope. Audit 3318 composes adopted sources 3310, 3313 and 3317: intentionally retires 71 authored declarations and 102 expanded cases, including whole terminal-detach agent-retention, seeded batch/single open-files equivalence, shallow context-menu shortcut-badge, mocked active-workspace delete-dispatch, workspace session hydration, SSH target cleanup, editor session hydration and worktree terminal purge suites. Lost alarms include chained transfer alias and immediate close/exit retention; batch/single divergence, survivor/selection/map preservation and input nonmutation across 400 imperative seeds (one registered case); disconnected pending-handle retry, authoritative absences/close-record shortcut/older-host absence and published-presence/unpublished-frame examples; published empty host-graph omission to pane exit; legacy search ranking/exclusion fallback; provider request upgrade/dedupe/finalizer ownership; asynchronous PR suppression/primary-change and same-number different-repository URL rejection. Deletion routing loses host/instance qualification, paired and legacy mapping, folder ownership, active-versus-hover/menu-row selection, ambiguity, modal/editable/creation/view/main/deleting/missing guards and shortcut advertisement. Unchanged real keyboard/hover/confirmation/catalog E2E remains narrower, not an equivalent replacement. Additional losses include scripted old-host disk/UTF-8/noop/baseline and landed/unlanded lost-reply/replacement-retry outcomes; ordinary/repeated healthy-link reattach; cross-owner terminal/chat binding; oversized replay permanent-unavailable classification; floating reveal keyboard ownership/close/disable/listener disposal; naming/retained-turn/Sleep/error/close/reject status tables; SSH membership prune/new-target-only reads and environment/local bucket isolation; same-frame receipt capacity and positive bare/headless publisher readmission. These actual regression alarms are deliberately lost; no fresh gate, provider, rendered, mobile, mixed-version, cross-platform, soak or measured-budget validation is claimed. Audit 3312 composes adopted sources 3302 and 3308: deliberately retires 33 authored declarations and 36 expanded cases, including the whole backend startup-focus, active-browser IPC-close-routing, remote-browser-creation and agent-terminal-presence suites. Intentionally lost alarms cover distinct six-pane reconnect handles; SSH host label/Connect callback, unhydrated visit cleanup, repo cache aliases and hidden worktree/tab/PTY cleanup; adopted-agent startup focus and viewer navigation/reuse; staged, pageless, source, stale and orphan browser-close routing; web local-materialization refusal, desktop incapable-host fallback, failed/thrown remote creation and paired-web sign-in refusal; BOM/NUL and remote/local custom-editor byte-read examples; immediate folder activation and unread clearing; and batching idle/coalescing, accumulated byte ceiling, leading-edge-disabled, standalone synchronous oversize rejection, validation yield, deferred pending-tail order and both viewport-clear examples. The newly reintroduced agent-terminal-presence suite also loses actual terminal-status membership positive and nonterminal-status callback suppression alarms, including the recent 40916 guard regression; the older different 2133 presence-suite history remains historical and receives zero duplicate removal credit. These real regression alarms are intentionally lost. The unchanged survivors provide narrower coverage, with no equivalent replacement or fresh gate, provider, rendered, remote/mobile, mixed-version, cross-platform, soak or measured-budget validation claimed. Audit 3301 combines adopted source batches 3296 and 3299: retire 27 authored declarations and 36 expanded scenarios, including six whole suites. Intentionally lost alarms cover failed-verification/recovery/restart counters and mocked hosting/toast; early overlapping pane-read scheduling/count; unreadable removal listing draft guard; bulk-purge remote-qualified and raw-local guard maps; initial-snapshot blank/mode/oversize/printable/fresh-spawn callback and replay examples; fallback mobile file close and ordinary MRU; default queue overload/drain; seeded browser-page reuse and supplied missing-page URL recreation; ambiguous sleep/hibernation owner refusal and untouched bindings/origins; ordinary selection/activity/fallback priority; skip-remote success and retained hints; supplied structured-chat unsent retention and scoped removal draft eviction; caller-chat focus precedence; open/draft PR cache eviction and merged-head fallback result inventories. Retained adverse-ordering and ownership cases are narrower and do not establish equivalent replacements. No fresh runtime, native, physical provider, cross-platform, mobile or mixed-version evidence is claimed. Audit 3292 combines adopted 3285/3287/3289/3293 removals: 40 authored declarations and 47 expanded scenarios, including whole frame-editor-view, ambiguous remote-delete-refusal, unverifiable-host subscription and editor-close cleanup suites. The previously reviewed dock-badge suite was already removed upstream in PR 40895 and receives zero new removal credit. Intentionally retired alarms include background browser selection and last-tab landing; main startup fallback, SSH project cleanup/survivor routing, exact-PR contradiction/saved-push positives, replay/sidecar rollback and wrapped remote error normalization; records/snapshot missing-diff hydration, ambiguous folder-owner delete refusal, focus no-op notifications, nested editor context-only cache check, pending broadcast-first status and ordinary four-confirmation aggregation; preview-source frontmatter pruning, approved local/runtime agent dispatch, non-GitHub legacy PR cache invalidation, stale merged-head transient-error fallback, empty-queue immediate query reply and unverified-to-concrete preflight risk. Additional intentionally retired alarms are injected host subscription/unsubscribe counts, synthetic editor reveal/draft/fallback/scoped close, ambiguous unread refusals and hidden mark/read roundtrip, immediate connect backoff/noError, positive mirrored evidence-clock renewal, supplied VM failure/success projections and git-local/folder-SSH wake permutations. Remaining temporal/ownership cases are narrower; no equivalent replacement or fresh execution is claimed. Audit 3283 removes 20 authored declarations and 31 expanded scenarios, including two whole suites. Retired alarms cover PR refresh-state pruning, shared runtime subscription cleanup and store isolation, language content-change routing and older completion ownership, ordinary panel retry-timer cleanup and pending loading, four hosted-review lifetime permutations, mocked terminal creation error/order wiring, five sleep-inventory variants with pending-frame-only assertions, palette fallback/composite highlights, Windows/UNC encoded-editor input rows, mixed-repository partial failure counts, title setters and generated-title bypass/replacement, five window-close coordinator delegation/guard/finally branches, and ambiguous-old-record frame reveal. These regression checks are intentionally lost under the test-value audit; retained cases are not equivalent replacements. Prior coverage notes and receipts remain historical. Audit 3273 removes 14 authored declarations and 19 expanded scenarios across four whole suites and eight partial suites. Retired alarms cover saved-local cancellation/layout ownership, ambiguous-owner warning reset, folder attachment routing, VM false-capability bypass, abandoned reveal warmup cleanup and viewport claim deduplication/rearming, quiet unrelated remote-removal refusal, startup-command one-shot consumption, repository/floating recovery-ledger rows, fake-React pinned-scroll intent, split-ratio state identity/subscriber suppression and fractional change, cross-host project-group join refusal, three PR request-lifetime permutations, upstream-status preservation on error, and unresolved Git-owner refusal. These regression checks are intentionally lost under the test-value audit; retained scenarios are not equivalent replacements. Historical coverage receipts remain historical, including gate35 subscriber claims. Audit 3256 removes these specific alarms from the broad root unit commands: skill dialog active-share selection, close and pending-share handoff; simple keyed-record deletion; marker live-region insertion, simple landing and nine baseline policy examples; General search/control navigation inventory; orphan-count cache bootstrap; seven Codex code classifications and HTTP 429 warning mapping; two skill unread-root verdict examples; peer-mail payload and conversation-ID rejection; and the GitHub organization TABLE text-field optimistic rollback/failed-envelope example. These include real behaviors; retained tests do not provide equivalent coverage for every removed scenario. Mobile losses are recorded separately and are not coverage of these root commands. Audit3246 retires NewRowsMarker reveal/no-repeat and pane-scoped beforeinput checks; skill-owner option availability/count/dedup/order; six styled Input/Textarea IME forwarding rows; the static Codex/OpenCode smart-key inventory; tab-map AST ratchet diagnostic fixtures; child-roster work-budget and retained-aggregate alarms; malformed persisted floating-cwd fallback/no-crash/empty exact-root checks; the unanswered-host empty-list discriminator; the direct invalidated-cache commit-false assertion; one redundant digit-keyup IME row; cached non-supported Claude title-generator rows; and the duplicate late naming-result singleton. These root broad commands lose those specific alarms without equivalent replacement. Distinct retained owner/state checks have their own bounded claims; the real scheduler still rejects stale response data, while the direct false-return alarm is retired. Audit3231 retires supported-host restore-pane key mapping, closed/open creation-sidebar toggle visibility, unrelated scope-error/null classification and quick-submit reveal/no-create checks. These broad commands lose those real narrow alarms; no replacement coverage is claimed. Audit3219 removes the synthetic Cmd+J extra-source suite; this broad command no longer checks its match/tie placement and selection, announced count, empty-query cap, or workspace hiding with retained recent tabs. No replacement coverage is claimed. Audit3209 composes approved sources3202/3206: deliberately retires seven authored/expanded cases in four whole suites without equivalent replacement. Work-item queries lose real GitHub/GitLab oversized ASCII secret-prefix empty-query and whitespace-before-trim refusal alarms; inputs share the imported size bound and exercise only its fast length branch, with no multibyte or actual provider/exposure guarantee. Skill sharing loses complete/partial managed-bundle and destination-ambiguity association checks plus three negative origin/path/scheme admission examples; these do not establish uploaded contents, arbitrary-URL fetching, authorization or execution. Retain exact originless-project direct-link fallback, cloud-delete confirmation, bundled-target repair/owner isolation and explicit Windows-host preference over stale WSL suites. Rejected3201 runtime-preference removal is excluded. No real cloud/network/host/disk, mobile or mixed-version execution validation is claimed. Audit3200 composes approved source3194: deliberately retires thirteen authored registrations/twenty-seven expanded cases in six whole automation host/list policy suites without equivalent replacement. Real RPC category/retryability and wrapped-timeout recognition, unscoped-failure/scoped-success health evidence precedence, exact-ID create selectors and per-run workspace omission, paired-web local/SSH/absent host catalog routing, incompatible/removed/unhydrated/offline execution health, runtime-before-nested-SSH recovery dispatch, and no-owned-focus/successor/foreign-list focus policy alarms are consciously lost. Retain exact20line orphan allow/refuse/precondition and exact90line cross-host same-ID search suites. No retry-timer/fence lifecycle, actual reconnect/network/write delivery, native focus, mobile/mixed-version runtime or disk persistence coverage is claimed. Audit3191 combines approved source3182/3187/3188: deliberately retires thirteen authored Vitest registrations and fifteen expanded cases, including five whole suites, without equivalent replacement. Attachment intake loses real portal/native-capture terminal interception and unowned-container refusal, immediate child-only committed availability/claim-ref negotiation and delivery, and direct-target over stable focused/background composer routing through disable/re-enable/remount and independent question focus. Exact destination/cache/draft and final-stat owner/workspace/File-release suites remain; they do not replace Surface recipient routing. Automation authoring loses destination runtime catalog membership, local/direct-SSH exclusion of paired-runtime repos, unknown-runtime empty catalog, logical project/host/setup/repo/path context payload, and missing/nonready setup refusal; these were genuine direct helper alarms, while exact schedule transformation, legacy schedule and setup-decision suites remain. Agent picker loses setup-chat/terminal/missing-installer/unsupported-or-unknown-host row projection samples and the ordinary unsearched menu variant; the entire original26-line resume recipe suite remains, including reported bypass versus default/absent permission modes, as does the exact search-result menu callback with its fixtures, model/effort commits and staged Arguments reopen outcome. Rejected predecessor3180 full recipe deletion is excluded. No actual native drag/cursor/focus/IME, upload bytes or disk/restart persistence, full automation save/scheduler/runtime transport, real resume-process flags/environment delivery, or native menu/rendered layout validation is claimed. Audit3178 combines source3169/3170/3175: deliberately retires eight authored units (six Vitest cases and two imperative custom relay scenarios), including two whole suites, without equivalent replacement. Voice Section loses per-model pending-delete isolation across interleaved A/B completion and the two-success refresh-call alarm; exact VoicePane delayed key-clear/newer enabled-toggle merge remains and does not replace row deletion coverage. Speech loses dedicated rating-scale/inclusive-ceiling, unsupported-versus-unmeasured, explicit/system-language precedence/locale alias, accuracy versus fast named/default/best-measured selection, and unmeasured/no-match examples; exact offered-user-pick versus retired-selection case and its fixture remain. Relay loses legacy ping-delivery count/liveness sampled before the silence deadline and grouped malformed Bearer/host-ID 401/429 routing checks; twelve real Worker/DurableObject/shipped-client scenarios remain, not equivalent replacement for those narrow alarms. Six prior speech cases are not all deleted: approved successor3175 keeps the original user-choice case. No actual credential/file persistence, native/GPU/engine/download delivery, production assignment policy or deployed-runtime validation is claimed by this reduction. Audit3165 combines source3160/3162: deliberately retires six authored/expanded cases across three whole suites without equivalent replacement. Launch/status losses: observed React-render pending/resuming coordination across settings and deferred offer read; stale-offer reread suppressing dialog reopen; held-local-chat recovery visibility with new-chat preferences off; disabled-local admission avoiding offer RPC despite an injected but unconsumed paired-tab fixture; and run-only lost-reply uncertainty reachability until direct viewed-run retirement. Launch probe does not prove arbitrary synchronous-store atomicity, the paired fixture is not remote runtime enrollment, and the status case directly marks viewed state rather than rendering the summary. Crew-hook losses: old-host method_not_found cache persists across party changes but is released by independently supplied connection generation/verified pairing status or pairing revision, with actual frame publication and teardown. These are real narrow coordination, admission, uncertainty and mixed-version cache alarms consciously lost. Exact web verified-runtime writer handoff guard remains; it does not replace those alarms or prove live host/native/mobile delivery. Audit3156 combines3147/3150/3152/3153/3154: deliberately retires39authored/40expanded native-chat recovery cases, eight whole suites and two orphan test-support files without equivalent replacement. Lost alarms: lost-reply unconfirmed reconciliation, skipped rows/counts and missed-frame repair, unknown/pending outcome and omitted-failure-list compatibility, uncertain history after dismissal, retry/refusal/dismiss lifecycle, hidden/consumed picker ownership, result-tree keyboard focus/disclosure and all-skipped retirement; unconfirmed-over-conflict guidance and failed-chat changes despite older host timestamps; nearest-listed ancestor/cyclic forest shape, descendant membership, host ordering/missing-host local fallback; skipped/readiness/admission counts, provisional skip/progress monotonicity, early per-chat verdicts with batch pending protection, older-host waiting, start-slot/delivery distinction, retry-over-failure and unknown-after-withdrawal attention; remote snooze/reconnect/reoffer, unseen mixed summary through open/close, watched completion retirement, new success beside old failure, failed-reread mixed-result preservation, filter/collapse completion resets, retry summary replacement and current-offer/history selection; cross-host workspace identity, rendered cyclic reachability, nested tri-state/toggle propagation, project eligibility isolation and collapse preserving selection. Exact fallback-list completion barrier, batched verdict/cleanup retention, stale-read versus dismissal publication race, and requested-but-never-rendered finished-run retirement remain; they do not replace retired alarms. Audit3146 combines source reviews3132/3133/3135/3138/3143: intentionally retire20authored/31expanded cases without equivalent replacement. Provider-session losses: structured-native terminal-resume exclusion, metadata-less completed identity, new-turn completion clearing, Pi provider-only/periodic/manual promotion, changed-transcript launch-config isolation, Pi/omp/prime-agent completed-session remote/manual-shutdown promotion, and quit/done precedence over heartbeat. Side-thread/question losses: finished-pane acquisition, hidden title/recap states, local/relay/relay-done conversation takeover with prompt/session/child resets, plain-message abandonment, compaction/session replacement and completion resets. Codex startup losses: incoming-owner cached child replacement, old-relay foreign live-session correction, completed foreign-root turn-fact permutations. Transcript losses: real temporary-JSONL partial sibling/final-child completion, unlabelled child attribution, compact cached replay preservation and100/1000completed-child roster retention/release. Compiler-ratchet losses: valid-line uncompiled-list parser grammar examples and injected-existence stale-path filtering; these did not exercise actual filesystem canonicalization or build exclusion. These are real lifecycle, ownership, replay, liveness and retention alarms, including recent#40246 regressions, deliberately lost. Retained conditional record replacement, clockless root/SSH owner isolation, persisted main-thread fork rejection, partial/final question-answer lifetime, cached-roster/missing-ID/live HTTP/durable startup boundaries and restarted-relay refusal/acceptance do not replace those losses. No native/SSH/WSL execution or equivalent coverage is claimed. Audit3128 scoped coverage reduction: Audit3121 removes VoiceDictationLanguageSetting.test.tsx (Spanish-to-es callback and unsupported-model Korean option presence) and voice-speech-model-rows.test.ts (provider/language-order/retired-on-disk list policy); these real narrow alarms are deliberately retired without replacement. VoiceSpeechModelSection per-model concurrent deletion remains unchanged. Audit3122 removes the direct absolute/parent-segment sparse-directory exception-type sample and the five-row IPC sparse-preset error-code/non-save table (two authored/six expanded cases). Those real typed-validation and IPC refusal/error-code alarms are deliberately retired. Keep both real symlink/junction floating-directory trust cases and the minimal RPC structured validation-reason serialization case. The RPC case does not replace IPC save/refusal behavior, typed normalization for both path inputs, actual Git sparse checkout or rendered localization. No equivalent replacement or fresh test/runtime/platform/remote/performance validation is claimed. Audit3124 removes ExperimentalPane.test.tsx (ordinary Pet switch callback), WorkspaceDirectorySetting.test.tsx (padded absolute saved-path normalization/no-write blur), and repository-search.perf.test.ts (navigation/selected-pane builder-call policy, fixture cardinality/nonempty entries and renamed project keyword inclusion). These real narrow alarms are retired without replacement; RepositoryPane.search.test.tsx search-to-sparse-editor request remains unchanged. Audit3129: resolver Windows case-insensitive and POSIX case-sensitive configured-env composition plus mocked-base launch-preflight Windows composition are intentionally no longer checked by the two deleted suites. No equivalent coverage claimed. Keep runtime-structured-agent-host-env.test.ts unchanged: configured Windows Claude-home account identity and host-account precedence remain checked. Audit3114 incorporates source3101: retire three structured Claude child-environment wrapper cases covering inherited stamp isolation/default and explicit entrypoint, Windows SDK-key canonicalization/default/duplicate-key order, and configured lookup/final environment/config-dir conjunction. Keep direct auth/session sanitation and Windows-versus-Linux overlay checks, actual executable lookup from supplied inherited PATH, native Windows child environment delivery, and stale option-mutation fencing. These are not equivalent replacements for the retired wrapper alarms. No fresh test/runtime, real login-shell probe, Claude SDK startup, remote/mixed-version, UI or performance result is claimed. Audit3114 incorporates source3105: retire 17 authored/20 expanded daemon launch-note cases, including the entire two-case SessionLaunchNote suite. Deliberately lose split-marker carry across same-token re-arm, stale-token retirement refusal, foreground tri-state and serial settlement, omitted-note/no-incarnation mixed-version policy, sticky retirement, first-incarnation/replacement launch preservation and held-launch precedence. Keep only the unchanged unresolved foreground-read race against a later finished attach. Supplied notes/foreground and runtime calls do not establish actual daemon/main restart, SSH, native Windows observation, mobile UI, equivalent replacement or fresh runtime/performance validation. Audit3114 incorporates source3109: retire six authored/35 expanded pricing cases: 21 accepted model-name examples, 10 unsupported/future-model fail-closed examples, malformed TTL clamp, unknown-model estimate refusal, flat-model long-context disregard and oversized long-context clamp. Keep mixed five-minute/one-hour write allocation and short/all-long/mixed long-context allocation with original fixed oracles. Real parser/refusal/bounds alarms are knowingly lost without equivalent replacement; retained tests do not validate provider rates, actual billing or request threshold classification. Preserve the archived usage-scan-append source-freeze receipt without transferring historical execution/performance credit. No fresh runtime result is claimed. Audit3108 source-only reduction (3094/3095/3100): retire 27 authored/30 expanded cases, including complete markdown view-state, file-preview link-open and CLI agent-server-route suites, without replacement. Deliberate losses include preview visibility/reference lifetime and same-path owners; SSH stat, directory/missing-target, explicit-null owner, nonmarkdown line/column, local URI routing, remote-to-client guard and same-file reuse; document-preview source-owner/folder/reuse routing, stable grant location and outside-workspace POSIX/Windows request projection; and verified remote CLI send/check refusal, suggested --on argv/copy, shell no-probe bypass, unverified-launch admission and older-peer refusal. Retain two markdown source-runtime/qualified-reveal alarms, one stale grant-rejection fence and two remote read/mutation evidence policies, with original callbacks. Final last-wins reveal did not prove transient cancellation; CLI prerequisite did not prove no real send, and envelope tests do not execute server refusal. No fresh runtime, rendered UI, actual file/SSH/paired transport, authorization, mixed-version or performance result is claimed. Audit3096 source-only reduction (3087/3088): retire 69 authored/76 expanded cases, including complete recovery-portals and reconnect-banner suites. Intentionally lose dead/default TUI replay, generated alternate-frame style composition, reset/mode/dimension and legacy missing-mode-boundary alarms, UI notice/action/reveal policies, and isolated recovery lifecycle variants, without equivalent replacement. Retained registrations and callbacks are unchanged; headless cells are not pixels, and no fresh runtime, host, SSH, WSL, rendered-app or performance result is claimed. Audit3086 source-only reduction (3079/3080/3081): intentionally retire 93 authored/100 expanded mock-boundary cases, including the complete spawn-errors and SSH reconnect overlay suites, without replacement. Retired alarms include spawn classification/ownership, reply ceilings and renderer-write failure recovery, disconnect/detach fencing branches, authority facts, and overlay retry/visibility wiring. Retained callbacks are unchanged; no fresh execution, real PTY/SSH, rendered UI, or performance result is claimed. Audit2432: The broad unit command loses the isolated BranchPicker open-change notification sample. Its unique true/false callback alarm is intentionally retired; reopening/status-refresh behavior was not exercised. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2425: The broad unit command loses the workspace-copy-status supplied readError forwarding sample. It did not exercise the branch-switch blocking promised by its title. This alarm is intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2417: The broad unit command loses two ordinary clean Git cherry-pick/revert success samples and two terminal startup foreground-read work-budget samples. These unique alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2412: The broad unit command loses the whole six-case fake-React-hook scroll-anchor suite, two generic draft-retention samples, a simple new-agent order sample, a real history-publication batching-count sample, an unmounted handoff abort/timer sample and three basic WSL preview/error samples. These alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2404: The broad unit command loses the standalone share-name lookup 503 propagation suite, initial deletion placement-kind sample, direct changed-mtime stale sample, first-pass rollback existence example and abandoned-preparation directory sample. These exact alarms are intentionally retired without equivalent replacement or fresh validation of the reduced selector. Audit2397 review amendment: The broad unit command loses three standalone suites, two partial-case definitions and two URL table rows across package/download/placement tests. Keep initial wrong-port authorization, modified independent-copy digest refusal and abort-after-begin cancellation dispatch. The remaining alarms are deliberately retired; no equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2393 current-parent refresh2401: Remove the basic skill-install service alias sample and original public managed-remove receiptless refusal case. Preserve the upstream request ownership suite and transaction interruption coverage. No equivalent complete replacement or fresh validation of the reduced selector is claimed. Audit2389: The broad unit command loses the missing-preview suite, plain-global destination sample and prepublished-receipt recovery example. No equivalent replacement or fresh validation of the reduced selector is claimed. Audit2385: The broad unit command loses four authored walker/staging/protocol examples (six expanded cases); no equivalent replacement or fresh validation is claimed for the reduced suite. Audit2383: The unscoped Vitest command no longer includes the removed Antigravity catalogue suite; its exact provider-root/owner example has no equivalent replacement or fresh validation claim. Audit2377/2381: The unscoped Vitest command includes the reduced bundled-skill and native-discovery suites; the eight removed backup/legacy/catalogue/symlink/SSH-builder examples have no replacement-equivalence or fresh validation claim here. Audit2133 current-parent preparation: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Audit2095 current-parent preparation2126: Current reduced sources/selectors intentionally retire the alarms listed in knownGaps; prior dated runs and replacement claims do not validate the changed set. Historical prior account (preserved verbatim): Require both existing named browser placement scenarios to pass three times with one attempt, zero skips, zero failures, and no report errors.", "commands": [ "gh workflow run packaged-browser-e2e.yml", "pnpm exec playwright test tests/e2e/packaged-mixed-version-browser-placement.spec.ts --config tests/playwright.config.ts --project=electron-headless --workers=1 --repeat-each=3 --retries=0", @@ -14170,7 +14177,9 @@ "Audit3446 composes ROOT-adopted3441/independent3442: intentionally removes100 authored registrations/119 potential expanded cases (118 conditional-active before platform gates) across16 sources, retaining11 authored/13 expanded cases and deleting6 whole suites. Two parent registrar import/call cleanups and the unused68line CLI recording fixture carry zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit and mobile discovery loses interrupt/wake registrars, task storage/session-close/account snapshot hooks, permission tables, copy/preference/model-discovery/composer examples, launch placement examples, VM projections and extra mobile-history scenarios. Retained checks cover abandoned versus dispatched fork uncertainty; /fast dispatch supersession; discovery lock release after exit; stale-send draft preservation; OSC status lifetime; future-version file preservation; launch cancellation before/during dispatch on both layout owners; mobile resume addressability before hooks; and recipe snapshot cleanup after plugin removal. The paired browser-link Electron source is unchanged. Terminal launch seams, mobile roles-only history and injected provider results do not establish native terminal destruction, transcript text fidelity or real remote provider cleanup. All prior notes and complete dated receipts retain their original scopes.", "Audit3451 composes ROOT-adopted3444/independent3445: intentionally removes101 authored registrations across16 reviewed sources, deleting7 whole suites and reducing8; one reviewed source is unchanged. Fresh main has123 authored registrations before and22 after because an upstream direct-child shell regression remains verbatim with zero removal credit. Registered removals are100+7*S and explicit-active removals99+7*S, where S is the unprobed count0..3 of admitted bash/zsh/fish shells. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses extra stream lifecycle/forwarding examples, unreadable-store/cache projections, orchestration presentation/benchmark cases, remote-owned routing matrices, mobile legacy navigation and model-catalog listing/cache examples. Whole suites retire renderer client-event projections, manually assembled structured-session reader ownership, mobile prompt presentation, renderer AI-vault resume commands, mobile Jira URL lookup, composer retention and mobile folder source-repository queries. Retained checks preserve original stream/store/orchestration/routing/navigation/catalog cases and shell behavior, including the upstream direct-child case. No real provider, native desktop or mobile acceptance is established by this reduction. All prior notes, assertion references and complete dated receipts retain their original historical scopes.", "Audit3456 composes ROOT-adopted3449/independent3450: intentionally removes62 authored/82 potential expanded cases from16 selected drivers, including7 whole suites, while13 authored/13 potential cases remain. Delete the unused48line legacy workspace-session JSON fixture with zero test credit. These are source-only reductions without equivalent replacement or fresh runtime claims. Broad unit discovery intentionally loses whole trust-dialog observer, background-turn notifications, redirected session refusal, typed-agent identity, chat-tab lifetime-cost, model-manager and mobile diff-review controller suites, plus individual persistence, synthetic composition, pairing, manifest and subprocess samples. Retain deferred snapshot dirty-domain persistence, two direct editor ownership cases, stream counter-domain distinction, two persisted pairing identity cases, parser entrypoint asymmetry, dictation retry and process-group timeout; paired restart and packaged old-binary/corrupt-profile journeys remain in their E2E driver scope. Retained checks do not replace removed alarms. No fresh execution, native IME/microphone, real authentication, mixed-version/platform or performance result is claimed. All prior notes and archived receipts retain original scopes.", - "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." + "Audit3468 accepted source proposal AUTHOR3462/PEER3463, composed privately by audit3469: removes62 authored registrations/95 finite potential expansions, including5 whole suites, retaining25 authored registrations/29 finite potential expansions. ROOT approved source value; no equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses the profile-state worker smoke, worker-agent views and three search discovery suites, plus samples from node-server qualification, skill-bundle generation, CLI hooks/profile/recovery/offline admission, permission-hook holds and search indexing. The separate Cloud relay runner loses one relay blackbox sample; unchanged mobile native-chat and dependent-transcript drivers remain. Retained source callback/table/ancestor/fixture bytes bind to the sealed candidate. No physical-device, mixed-version, remote, cross-platform or app execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes.", + "Audit3438 composes ROOT-adopted3434/independent3436: intentionally removes60 authored/62 potential expanded cases across16 sources, deleting9 whole suites. Eight authored/eight potential cases remain; delete the102line unused ACP fixture and57line unused teardown helper with zero test credit. No equivalent replacement or fresh execution is claimed. Broad unit discovery loses ACP model-catalog fixtures, resume-record projections, dashboard mock payload/state cases, mobile folder/launch/session examples and dictionary pipeline rows. Selected E2E removals also retire auto-conversion and golden-core journeys and one Windows CLI conversion registration. Retain ACP reply text across a resumed subagent, both VAD checks, focused tab send ownership, async focus ownership in happy-dom, the question-skip journey, the original managed Windows CLI lifecycle journey and dictionary substitution. These checks do not replace the removed alarms. All prior notes and dated/archived evidence retain original scopes.", + "Audit3470 composes ROOT-approved AUTHOR3466/PEER3467 source proposal: intentionally removes38 authored registrations/59 finite potential expansions, including8 whole registering suites, retaining13 authored registrations/18 finite potential expansions. The sole dead shell-recipe caller cleanup has zero additional test credit. No equivalent replacement, fresh runtime or merge credit is claimed. Broad root-unit discovery loses process-import boundary, Korean/Japanese locale policy, worker-reset, mobile agent-history/queued-message and median-summary suites, plus packaged mobile-manifest, worker-pool fixture, HTML preview, mobile shell and terminal samples. The environment-provider contract and its sole shell-recipe caller are deleted. Separate cross-version-wire discovery loses the large byte-budget fresh-snapshot/replay sample while retaining independent live same-revision note publication; the retained terminal stale-refresh and dirty HTML draft persistence checks do not replace removed alarms. The source4 unused start/stop fixture hooks are removed; its retained nested failed-import test remains byte-exact with its own lifecycle. Unchanged SVG preview and Git-resolution registrations stay intact. Root balanced CI excludes the cross-version directory, which has a separate job; unbalanced root discovery can include its parent. No native app/device, actual remote, released-pair interoperability or cross-platform execution result is claimed. Earlier notes, complete dated receipts and archived gaps retain their original scopes." ], "demotionRule": "Keep experimental if any direction skips or fails; do not extend timeouts or retry to green." }, diff --git a/config/scripts/ci/sharding/ci-shard-timings.json b/config/scripts/ci/sharding/ci-shard-timings.json index 7c645830feb..48b7a195cc2 100644 --- a/config/scripts/ci/sharding/ci-shard-timings.json +++ b/config/scripts/ci/sharding/ci-shard-timings.json @@ -18,7 +18,6 @@ "config/scripts/static-analysis/oxlint-contracts/app-store-performance-plugin.test.mjs": 339, "config/scripts/desktop-distribution/linux/appimage-node-runtime.test.mjs": 146, "config/scripts/performance/measurement/benchmark-artifact-comparison.test.mjs": 192, - "config/scripts/performance/measurement/benchmark-sample-summary.test.mjs": 46, "config/scripts/cli-build/build-cli.test.mjs": 1371, "config/scripts/desktop-distribution/linux/build-linux-local.test.mjs": 67, "config/scripts/desktop-distribution/macos/build-mac-notarization-overlap.test.mjs": 127, @@ -100,9 +99,7 @@ "config/scripts/locale-catalog/locale-brand-prefilter.test.mjs": 246, "config/scripts/locale-catalog/locale-generic-ui-terms.test.mjs": 102, "config/scripts/locale-catalog/locale-repair-catalog-missing-leaves.test.mjs": 159, - "config/scripts/locale-catalog/locale-translation-policy-ko-round5.test.mjs": 154, "config/scripts/locale-catalog/locale-translation-policy.es-pr.test.mjs": 91, - "config/scripts/locale-catalog/locale-translation-policy.ja-relocalization.test.mjs": 116, "config/scripts/locale-catalog/locale-translation-policy.test.mjs": 189, "config/scripts/localization/source-extraction/localization-extraction-call-syntax.test.mjs": 317, "config/scripts/localization/extraction-policy/localization-extraction-change-scope.test.mjs": 732, @@ -115,7 +112,6 @@ "config/scripts/orcad-distribution/prebuilds/merge-orcad-prebuilds.test.mjs": 67, "config/scripts/mobile-web/check-entry/mobile-process-host-tooling.test.mjs": 942, "config/scripts/mobile-web/check-entry/mobile-release-check-scope.test.mjs": 821, - "config/scripts/mobile-web-app/session/mobile-web-app-agent-history-render.test.mjs": 524, "config/scripts/mobile-web-app/shell/mobile-web-app-agent-icon-render.test.mjs": 591, "config/scripts/mobile-web-app/shell/mobile-web-app-browser-pane-render.test.mjs": 239, "config/scripts/mobile-web-app/shell/mobile-web-app-clipboard-image-resize-render.test.mjs": 422, @@ -133,7 +129,6 @@ "config/scripts/mobile-web-app/routing/mobile-web-app-page-closure-families.test.mjs": 71, "config/scripts/mobile-web-app/routing/mobile-web-app-page-grant-call-sites.test.mjs": 164, "config/scripts/mobile-web-app/routing/mobile-web-app-push-subsystem-closure.test.mjs": 57, - "config/scripts/mobile-web-app/session/mobile-web-app-queued-messages-render.test.mjs": 267, "config/scripts/mobile-web-app/shell/mobile-web-app-render.test.mjs": 254, "config/scripts/mobile-web-app/session/mobile-web-app-rich-markdown-render.test.mjs": 356, "config/scripts/mobile-web-app/mobile-web-app-route-manifest.test.mjs": 174, @@ -460,7 +455,6 @@ "src/main/acp/process/acp-agent-connection.test.ts": 407, "src/main/acp/discovery/acp-at-rest-commands.test.ts": 1400, "src/main/acp/transport/acp-json-rpc-peer.test.ts": 195, - "src/main/acp/discovery/acp-model-catalog-probe.test.ts": 774, "src/main/acp/protocol/acp-permission-requests.test.ts": 254, "src/main/acp/turns/acp-prompt-content.test.ts": 302, "src/main/acp/protocol/acp-session-agent-turns.test.ts": 260, @@ -4543,8 +4537,6 @@ "src/renderer/src/components/automations/hosts/use-automation-host-catalog-readoption.test.tsx": 802, "src/renderer/src/components/automations/hosts/use-automation-host-catalog.test.tsx": 920, "src/renderer/src/components/automations/listing/use-automation-list-search.test.tsx": 679, - "src/renderer/src/components/automations/runs/use-automation-runs-dashboard-payload.test.tsx": 6692, - "src/renderer/src/components/automations/runs/use-automation-runs-dashboard.test.tsx": 854, "src/renderer/src/components/automations/use-selected-automation-run-history.test.tsx": 2104, "src/renderer/src/components/browser-favicon.test.tsx": 333, "src/renderer/src/components/browser-pane/ClientHostedBrowserPagePane.drag-focus.test.tsx": 5826, @@ -7361,7 +7353,6 @@ "tests/e2e/helpers/electron-process-shutdown.unit.test.ts": 6545, "tests/e2e/helpers/golden-source-control.unit.test.ts": 804, "tests/e2e/helpers/headless-paired-runtime-serve-readiness.unit.test.ts": 53, - "tests/e2e/helpers/orcad-convert-teardown.unit.test.ts": 76, "tests/e2e/helpers/paired-client-runtime-environment.unit.test.ts": 67, "tests/e2e/helpers/paired-client-window-reveal.unit.test.ts": 57, "tests/e2e/helpers/persisted-profile-state.unit.test.ts": 402, @@ -7567,7 +7558,6 @@ "tests/e2e/github-cli-stall-repro.spec.ts": 18200, "tests/e2e/github-url-smart-input-transition.spec.ts": 58200, "tests/e2e/golden-agent-tui-launch.spec.ts": 27300, - "tests/e2e/golden-core-flows.spec.ts": 91100, "tests/e2e/golden-file-open-edit-save.spec.ts": 25000, "tests/e2e/golden-fresh-profile-terminal.spec.ts": 24400, "tests/e2e/golden-posix-fresh-startup.spec.ts": 4900, diff --git a/config/scripts/desktop-distribution/runtime-verification/verify-packaged-mobile-web-bundle.test.mjs b/config/scripts/desktop-distribution/runtime-verification/verify-packaged-mobile-web-bundle.test.mjs index e916dfc216a..74368296c74 100644 --- a/config/scripts/desktop-distribution/runtime-verification/verify-packaged-mobile-web-bundle.test.mjs +++ b/config/scripts/desktop-distribution/runtime-verification/verify-packaged-mobile-web-bundle.test.mjs @@ -8,8 +8,6 @@ import { writeMobileWebBundleFixtureTree } from '../../build/mobile-web-app/mobi const require = createRequire(import.meta.url) const { assertMobileWebBundleBuilt } = require('./verify-packaged-mobile-web-bundle.cjs') -const electronBuilderConfig = require('../../../electron-builder.config.cjs') -const { darkBuildMarkers } = require('../../../dark-build-switches.cjs') async function withBundle(run, { brotli = false } = {}) { const scratch = await mkdtemp(join(tmpdir(), 'orca-mobile-web-guard-')) @@ -37,16 +35,6 @@ describe('assertMobileWebBundleBuilt', () => { vi.restoreAllMocks() }) - it('fails on a file the manifest does not list, so no stale asset ships inside asar', async () => { - await withBundle(async ({ bundleDir }) => { - // Dropped content-addressed assets never overwrite themselves; reject leftovers rather than pack unverified unreachable bytes. - await writeFile(join(bundleDir, 'assets', 'stale.js'), '// from an earlier build\n', 'utf8') - expect(() => assertMobileWebBundleBuilt(bundleDir)).toThrow( - /does not list: assets\/stale\.js/ - ) - }) - }) - it('fails when an asset is tampered with on disk', async () => { await withBundle(async ({ bundleDir, manifest }) => { const asset = manifest.assets.find((entry) => entry.path.endsWith('.js')) @@ -58,14 +46,6 @@ describe('assertMobileWebBundleBuilt', () => { }) }) - it('fails when a listed asset was never written', async () => { - await withBundle(async ({ bundleDir, manifest }) => { - const asset = manifest.assets.find((entry) => entry.path.endsWith('.png')) - await rm(join(bundleDir, asset.path)) - expect(() => assertMobileWebBundleBuilt(bundleDir)).toThrow(/which is missing from/) - }) - }) - it('accepts the brotli copies the manifest lists, and fails one that inflates to other bytes', async () => { await withBundle( async ({ bundleDir, manifest }) => { @@ -84,29 +64,4 @@ describe('assertMobileWebBundleBuilt', () => { { brotli: true } ) }) - - it('refuses an asset path that escapes the bundle directory', async () => { - await withBundle(async ({ bundleDir }) => { - await rewriteManifest(bundleDir, (manifest) => { - manifest.assets[0].path = '../outside.js' - }) - expect(() => assertMobileWebBundleBuilt(bundleDir)).toThrow(/not a safe relative path/) - }) - }) -}) - -describe('electron-builder packaging wiring', () => { - it('verifies a dark page build, and refuses to package it', async () => { - await withBundle(async ({ bundleDir }) => { - const [marker] = darkBuildMarkers({ ORCA_SIDEBAR_INPUTS: 'main' }) - await writeFile(join(bundleDir, marker.fileName), marker.text, 'utf8') - expect(() => assertMobileWebBundleBuilt(bundleDir)).not.toThrow() - expect(() => - electronBuilderConfig.beforePack( - { electronPlatformName: process.platform, arch: process.arch === 'arm64' ? 3 : 1 }, - bundleDir - ) - ).toThrow(`${bundleDir} holds an ORCA_SIDEBAR_INPUTS=main build`) - }) - }) }) diff --git a/config/scripts/locale-catalog/locale-translation-policy-ko-round5.test.mjs b/config/scripts/locale-catalog/locale-translation-policy-ko-round5.test.mjs deleted file mode 100644 index 6d8561bcdcc..00000000000 --- a/config/scripts/locale-catalog/locale-translation-policy-ko-round5.test.mjs +++ /dev/null @@ -1,54 +0,0 @@ -import { describe, expect, it } from 'vitest' - -import { repairTranslatedValue } from './locale-translation-policy.mjs' - -describe('locale-translation-policy ko round 5', () => { - it('re-glues Korean particles after Latin terms without gluing content words', () => { - // A wrong space before the 에 particle is re-glued onto the Latin term. - expect( - repairTranslatedValue({ - key: 'auto.components.StarNagCard.cf82170065', - enValue: 'Could not star the repo. Make sure', - localeValue: 'repo 에 스타를 표시할 수 없습니다. 다음을 확인하세요', - locale: 'ko' - }) - ).toBe('repo에 스타를 표시할 수 없습니다. 다음을 확인하세요') - expect( - repairTranslatedValue({ - key: 'auto.components.github.project.ProjectCell.4b5b871da8', - enValue: 'No labels in this repo.', - localeValue: '이 repo 에는 라벨이 없습니다.', - locale: 'ko' - }) - ).toBe('이 repo에는 라벨이 없습니다.') - expect( - repairTranslatedValue({ - key: 'auto.components.terminal.pane.CloseTerminalDialog.6b9a6975f8', - enValue: - 'The terminal still has a running process. If you close the terminal, the process will be killed.', - localeValue: - 'terminal 에는 여전히 실행 중인 프로세스가 있습니다. terminal을 닫으면 프로세스가 종료됩니다.', - locale: 'ko' - }) - ).toBe( - 'terminal에는 여전히 실행 중인 프로세스가 있습니다. terminal을 닫으면 프로세스가 종료됩니다.' - ) - // A correct space before a content word that merely starts with a particle syllable is kept. - expect( - repairTranslatedValue({ - key: 'auto.components.JiraIssueWorkspace.76513c7898', - enValue: 'Close Jira issue preview', - localeValue: 'Jira 이슈 미리보기 닫기', - locale: 'ko' - }) - ).toBe('Jira 이슈 미리보기 닫기') - expect( - repairTranslatedValue({ - key: 'auto.components.GitHubItemDialog.3ab6ac0fc8', - enValue: 'Preview and edit the selected GitHub issue or pull request.', - localeValue: '선택한 GitHub 이슈 또는 PR을 미리보기하고 편집합니다.', - locale: 'ko' - }) - ).toBe('선택한 GitHub 이슈 또는 PR을 미리보기하고 편집합니다.') - }) -}) diff --git a/config/scripts/locale-catalog/locale-translation-policy.ja-relocalization.test.mjs b/config/scripts/locale-catalog/locale-translation-policy.ja-relocalization.test.mjs deleted file mode 100644 index 0ab813ce684..00000000000 --- a/config/scripts/locale-catalog/locale-translation-policy.ja-relocalization.test.mjs +++ /dev/null @@ -1,76 +0,0 @@ -import { describe, expect, it } from 'vitest' - -import { repairTranslatedValue, shouldPreserveEnglishValue } from './locale-translation-policy.mjs' - -const ja = (enValue, localeValue, key = 'auto.test.relocalization') => - repairTranslatedValue({ key, enValue, localeValue, locale: 'ja' }) - -describe('locale-translation-policy ja relocalization', () => { - it('leaves the second word of a two-word command in Latin', () => { - expect(ja('a commit hook or git commit fails', 'コミットフックまたは git commit が失敗')).toBe( - 'コミットフックまたは git commit が失敗' - ) - expect(ja('Run orca terminal', 'orca terminal を実行')).toBe('orca terminal を実行') - expect(ja('Install with pnpm install', 'pnpm install でインストール')).toBe( - 'pnpm install でインストール' - ) - }) - - it('leaves identifiers, flags, filenames, and Latin-only labels alone', () => { - expect(ja('Pass --agent to select one.', '--agent を渡して選択します。')).toBe( - '--agent を渡して選択します。' - ) - expect(ja('Edit agents.md', 'agents.md を編集')).toBe('agents.md を編集') - expect(ja('Agent SDK settings', 'Agent SDK の設定')).toBe('Agent SDK の設定') - expect(ja('Agent SDK settings', 'この Agent SDK の設定')).toBe('この Agent SDK の設定') - expect(ja('Run git commit -m', 'git commit -m を実行')).toBe('git commit -m を実行') - }) - - it('repairs translated code examples back to executable values', () => { - expect(ja('--model sonnet', '--モデルソネット')).toBe('--model sonnet') - expect( - ja("gh pr list --json number -q '.[0].number'", "gh pr list --json 番号 -q '.[0].number'") - ).toBe("gh pr list --json number -q '.[0].number'") - expect(ja('{basePrompt}', '{ベースプロンプト}')).toBe('{basePrompt}') - expect(ja('/pricing', '/価格設定')).toBe('/pricing') - }) - - it('keeps style blocks in English because a translated selector breaks the view', () => { - expect(shouldPreserveEnglishValue('.pp-card[data-card="starter"] .pp-cta')).toBe(true) - expect(shouldPreserveEnglishValue('@keyframes browser-flash { to { opacity: 1; } }')).toBe(true) - expect(shouldPreserveEnglishValue('.card { background: var(--card); }')).toBe(true) - // Why: the declaration requirement is what keeps prose containing braces translatable. - expect(shouldPreserveEnglishValue('e.g. {{value0}}')).toBe(false) - }) - - it('leaves ranges and token samples out of the ellipsis rule', () => { - expect(ja('Compare main...HEAD', 'main...HEAD を比較')).toBe('main...HEAD を比較') - }) - - it('preserves a selector that has no declaration block', () => { - for (const value of [ - 'div.pricing-grid > div.card.starter:nth-of-type(1) > a.cta', - '.pp-cta', - '#dialog', - '.pp-card[data-card="starter"] .pp-cta', - 'button.primary', - 'a:hover', - 'input[type="text"]', - 'localhost:3000' - ]) { - expect(shouldPreserveEnglishValue(value)).toBe(true) - } - // Why: two ordinary sentences carry two periods, which must not read as selector joins. - for (const value of [ - 'Open Settings > Git and try again.', - 'Show live workspace ports. Click it for workspace-scoped ports and external listeners.', - 'orca.yaml + local', - '.gitignore', - 'Status:', - 'Done.', - '#{{value0}}' - ]) { - expect(shouldPreserveEnglishValue(value)).toBe(false) - } - }) -}) diff --git a/config/scripts/mobile-web-app/preview/mobile-web-app-html-file-preview-render.test.mjs b/config/scripts/mobile-web-app/preview/mobile-web-app-html-file-preview-render.test.mjs index 1ae5168c591..79c6f18256e 100644 --- a/config/scripts/mobile-web-app/preview/mobile-web-app-html-file-preview-render.test.mjs +++ b/config/scripts/mobile-web-app/preview/mobile-web-app-html-file-preview-render.test.mjs @@ -124,30 +124,9 @@ async function framePixel(page) { }) } -const sourceText = (page) => page.getByText(MARKER, { exact: false }) - for (const engine of ['chromium', 'webkit']) { const describeRender = bundles ? describe : describe.skip describeRender(`an HTML file on the file-preview route on ${engine}`, () => { - it('paints the page, and Source shows its markup', async () => { - const { context, page, errors } = await openHtmlFile(browsers[engine]) - const frame = page.locator('iframe') - await frame.waitFor({ timeout: 60_000 }) - await expect.poll(() => framePixel(page), { timeout: 10_000 }).toBe(PAINT_RGB) - if (shots) { - await page.screenshot({ path: join(shots, `html-file-preview-${engine}.png`) }) - } - - await page.getByRole('tab', { name: 'View HTML source' }).click() - await sourceText(page).waitFor() - expect(await page.locator('iframe').count()).toBe(0) - if (shots) { - await page.screenshot({ path: join(shots, `html-file-source-${engine}.png`) }) - } - expect(errors).toEqual([]) - await context.close() - }, 90_000) - it('renders unsaved HTML edits and retains them when returning to Source', async () => { const { context, page, errors } = await openHtmlFile(browsers[engine], { editable: true }) await page.locator('iframe').waitFor({ timeout: 60_000 }) @@ -165,36 +144,5 @@ for (const engine of ['chromium', 'webkit']) { expect(errors).toEqual([]) await context.close() }, 90_000) - - it('opens a line-targeted file on Source, and a truncated one as text only', async () => { - const targeted = await openHtmlFile(browsers[engine], { line: '3' }) - await targeted.page - .getByRole('tab', { name: 'View HTML source' }) - .waitFor({ timeout: 60_000 }) - await sourceText(targeted.page).waitFor() - expect( - await targeted.page - .getByRole('tab', { name: 'View HTML source' }) - .getAttribute('aria-selected') - ).toBe('true') - expect(await targeted.page.locator('iframe').count()).toBe(0) - if (shots) { - await targeted.page.screenshot({ path: join(shots, `html-file-line-${engine}.png`) }) - } - await targeted.page.getByRole('tab', { name: 'Preview rendered HTML' }).click() - await expect.poll(() => framePixel(targeted.page), { timeout: 10_000 }).toBe(PAINT_RGB) - expect(targeted.errors).toEqual([]) - await targeted.context.close() - - const truncated = await openHtmlFile(browsers[engine], { truncated: true }) - await sourceText(truncated.page).waitFor({ timeout: 60_000 }) - expect(await truncated.page.locator('iframe').count()).toBe(0) - expect(await truncated.page.getByRole('tab').count()).toBe(0) - if (shots) { - await truncated.page.screenshot({ path: join(shots, `html-file-truncated-${engine}.png`) }) - } - expect(truncated.errors).toEqual([]) - await truncated.context.close() - }, 120_000) }) } diff --git a/config/scripts/mobile-web-app/session/mobile-web-app-agent-history-render.test.mjs b/config/scripts/mobile-web-app/session/mobile-web-app-agent-history-render.test.mjs deleted file mode 100644 index 81756c74935..00000000000 --- a/config/scripts/mobile-web-app/session/mobile-web-app-agent-history-render.test.mjs +++ /dev/null @@ -1,198 +0,0 @@ -import { mkdtemp, rm } from 'node:fs/promises' -import { tmpdir } from 'node:os' -import { join } from 'node:path' -import { afterAll, beforeAll, describe, expect, it } from 'vitest' -import { chromium } from 'playwright-core' -import { buildMobileWebAppBundle } from '../../build/mobile-web-app/build-mobile-web-app-bundle.mjs' -import { mobileWebAppDependenciesPresent } from '../mobile-web-app-bundle-dependencies.mjs' -import { - createBundleServer, - waitForPageFrames, - installShellDouble, - readBridgeFaultGrant, - readBridgePagePainted, - readBridgeProtocolVersion, - readShellCsp -} from '../preview/mobile-web-app-render-harness.mjs' - -// Browser-only history checks prove real RN-Web closure evaluation, corpus-backed rows, shell Back handoff and deferred chunk delivery. -// Domain file uses shared C1.10 harness to avoid line-cap conflicts between concurrent series. - -const HOST_ROUTE = '/h/render-check-host' -/** The pattern `init.pageRoutes` names, which is what the page matches a navigation against. */ -const HOST_ROUTE_PATTERN = '/h/[hostId]' -const SHELL_SESSION_ID = 'render-check-session' -const SHELL_BUILD_ID = 'render-check-build' -const SHELL_HOST = { - id: 'render-check-host', - name: 'Render Check Host', - endpoint: 'ws://render-check', - lastConnected: 1 -} - -const bundles = mobileWebAppDependenciesPresent() -const describeRender = bundles ? describe : describe.skip - -let scratch -let server -let browser -let origin -let cspHeader = null -let bridgeVersion = null -let faultGrant = null -let paintName = null - -beforeAll(async () => { - if (!bundles) { - return - } - cspHeader = await readShellCsp() - bridgeVersion = await readBridgeProtocolVersion() - faultGrant = await readBridgeFaultGrant() - paintName = await readBridgePagePainted() - scratch = await mkdtemp(join(tmpdir(), 'orca-mobile-web-app-agent-history-')) - const built = await buildMobileWebAppBundle({ outDir: join(scratch, 'bundle') }) - const served = await createBundleServer({ outDir: built.outDir, cspHeader }) - server = served.server - origin = served.origin - const executablePath = process.env.ORCA_MOBILE_WEB_RENDER_BROWSER - browser = await chromium.launch({ headless: true, ...(executablePath ? { executablePath } : {}) }) -}, 180_000) - -afterAll(async () => { - await browser?.close() - server?.close() - if (scratch) { - await rm(scratch, { recursive: true, force: true }) - } -}) - -/** A page carrying every signal these cases read: uncaught errors, console errors, script paths. */ -async function openPage({ shellRoute, shellGrants, shellPageRoutes = null, replies } = {}) { - const page = await browser.newPage({ viewport: { width: 390, height: 844 } }) - // Install channel at document start; entry reads it during evaluation, before load. - await page.addInitScript(installShellDouble, { - version: bridgeVersion, - sessionId: SHELL_SESSION_ID, - buildId: SHELL_BUILD_ID, - route: shellRoute, - host: SHELL_HOST, - storage: {}, - faultGrant, - // The harness falls back to the fault grant alone, which is the ungranted page. - grants: shellGrants ?? [faultGrant], - pageRoutes: shellPageRoutes, - replies - }) - const errors = [] - const scripts = [] - let reportUncaught = () => {} - const uncaught = new Promise((resolve) => { - reportUncaught = resolve - }) - page.on('pageerror', (error) => { - errors.push(`${error.name}: ${error.message}`) - reportUncaught(error) - }) - page.on('console', (message) => { - if (message.type() === 'error') { - errors.push(`console.error: ${message.text()}`) - } - }) - page.on('response', (response) => { - const path = new URL(response.url()).pathname - if (response.status() === 200 && path.endsWith('.js')) { - scripts.push(path) - } - }) - return { page, errors, scripts, uncaught } -} - -// Wait beyond entry mount for route content; deferred import chunk may still be arriving. -async function waitForRoute({ page, errors, uncaught }, route, awaitText) { - const named = (cause, what) => - new Error(`${route} ${what}: ${errors.join(' | ') || 'no page or console error'}`, { cause }) - const race = async (wait) => - Promise.race([ - wait.then( - () => null, - (error) => error - ), - uncaught - ]) - const cause = await race( - page.waitForFunction(() => document.documentElement.dataset.orcaWebEntry === 'mounted', { - timeout: 30_000, - polling: 250 - }) - ) - if (cause) { - const state = await page.evaluate( - () => document.documentElement.dataset.orcaWebEntry ?? 'absent' - ) - throw named(cause, `never mounted (entry ${state})`) - } - const paintCause = await race( - page.waitForFunction((needle) => document.body.innerText.includes(needle), awaitText, { - timeout: 30_000, - polling: 250 - }) - ) - if (paintCause) { - throw named(paintCause, `mounted but never painted ${JSON.stringify(awaitText)}`) - } - for (const fault of await page.evaluate(() => globalThis.__orcaRenderCheckFaults ?? [])) { - errors.push(`page fault: ${fault}`) - } -} - -/** Opens the document at `/`, the one path the shell serves, and lets the page route itself. */ -async function openRoute(route, awaitText, options = {}) { - const opened = await openPage({ shellRoute: { pathname: route }, ...options }) - await opened.page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, route, awaitText) - return opened -} - -/** Every grant-gated notify the page posted, whole and in order, as the shell received them. The - * paint report and the location report are not: every page posts them for its route screen. */ -async function readNotifies(page) { - const notifies = await page.evaluate(() => globalThis.__orcaRenderCheckNotifies ?? []) - return notifies.filter((frame) => frame.name !== paintName && frame.name !== 'location') -} - -describeRender('the agent-history route in a real browser', () => { - // Prove real RN-Web import/evaluation absent from mocked native units; no-RPC case paints failure, next corpus case paints a row. - // Resume/scope-tab interactions remain native C5.3 device coverage. - - it('hands its Back button to the shell, which no unit test can prove in a browser', async () => { - const route = `${HOST_ROUTE}/agent-history/wt-1` - const opened = await openRoute(route, 'Agent Session History', { - shellGrants: [faultGrant, 'navigate'], - shellPageRoutes: [HOST_ROUTE_PATTERN] - }) - await opened.page.getByLabel('Back').click() - await expect - .poll(() => readNotifies(opened.page)) - .toEqual([{ v: bridgeVersion, type: 'notify', name: 'navigate-back' }]) - expect(await readNotifies(opened.page)).toEqual([ - { v: bridgeVersion, type: 'notify', name: 'navigate-back' } - ]) - // Document has only replaceState entry; page-local Back would do nothing and look like a successful shell handoff. - expect(await opened.page.evaluate(() => location.pathname)).toBe(route) - expect(opened.errors).toEqual([]) - await opened.page.close() - }, 60_000) - - it('posts nothing for Back when the shell granted no navigate, and stays put', async () => { - // Withhold grant while preserving tap/handler/seam to prove handoff depends on permission. - const route = `${HOST_ROUTE}/agent-history/wt-1` - const opened = await openRoute(route, 'Agent Session History') - await opened.page.getByLabel('Back').click() - await waitForPageFrames(opened.page) - expect(await readNotifies(opened.page)).toEqual([]) - expect(await opened.page.evaluate(() => location.pathname)).toBe(route) - expect(opened.errors).toEqual([]) - await opened.page.close() - }, 60_000) -}) diff --git a/config/scripts/mobile-web-app/session/mobile-web-app-queued-messages-render.test.mjs b/config/scripts/mobile-web-app/session/mobile-web-app-queued-messages-render.test.mjs deleted file mode 100644 index 864d07093c3..00000000000 --- a/config/scripts/mobile-web-app/session/mobile-web-app-queued-messages-render.test.mjs +++ /dev/null @@ -1,202 +0,0 @@ -import { mkdir, mkdtemp, rm, writeFile } from 'node:fs/promises' -import { join } from 'node:path' -import { fileURLToPath } from 'node:url' -import { afterAll, beforeAll, describe, expect, it } from 'vitest' -import * as esbuild from 'esbuild' -import { chromium, webkit } from 'playwright-core' -import { - MOBILE_WEB_APP_NATIVE_PARITY_STYLE, - MOBILE_WEB_APP_ROOT_RESET, - mobileWebAppBuildOptions -} from '../../build/mobile-web-app/build-mobile-web-app-bundle.mjs' -import { mobileWebAppDependenciesPresent } from '../mobile-web-app-bundle-dependencies.mjs' -import { createBundleServer, readShellCsp } from '../preview/mobile-web-app-render-harness.mjs' -import { SUPPORTED_UI_LOCALES } from '../../../../src/shared/ui-locale.ts' - -const mobileDir = fileURLToPath(new URL('../../../../mobile', import.meta.url)) -const engines = [ - { - name: 'chromium', - launch: () => { - const executablePath = process.env.ORCA_MOBILE_WEB_RENDER_BROWSER - return chromium.launch({ headless: true, ...(executablePath ? { executablePath } : {}) }) - } - }, - { name: 'webkit', launch: () => webkit.launch({ headless: true }) } -] - -const entry = ` -import { useEffect, useState } from 'react' -import { createRoot } from 'react-dom/client' -import { View } from 'react-native' -import { GestureHandlerRootView } from 'react-native-gesture-handler' -import { MobileNativeChatQueuedMessages } from './src/session/native-chat/input/MobileNativeChatQueuedMessages' -import { MobileNativeChatComposer } from './src/session/native-chat/composer/MobileNativeChatComposer' -import { mobileQueuedMessageCards } from './src/session/native-chat/input/mobile-structured-queued-message-cards' -import { activateMobileUiLocale, translate } from './src/i18n/mobile-i18n' - -const [mode, locale] = location.hash.slice(1).split(':') -const texts = ['First queued message', 'Review the mobile transcript and keep every detail. '.repeat(30)] -function Queue() { - const messages = texts.map((text, index) => ({ - messageId: 'message-' + index, - position: index + 1, - state: 'waiting', - body: { kind: 'message', role: 'user', blocks: [{ type: 'text', text }] } - })) - return - true : undefined} - /> - -} -function Editor() { - const [ready, setReady] = useState(false) - useEffect(() => { - void activateMobileUiLocale(locale).then(() => { - window.actionLabels = { - cancel: translate('chat.composer.cancelEditing', 'Cancel editing'), - save: translate('chat.composer.saveEditedMessage', 'Save edited message') - } - setReady(true) - }) - }, []) - return ready ? - {}} - onSend={async () => true} - editSession={{ cancel: () => {}, refuseCommand: () => false, locked: false }} - sendSurfaceId="editor" - getSendCompletionGeneration={() => 0} - getComposerEditGeneration={() => 0} - onAttachImage={() => {}} - onMicPress={() => {}} - /> - : null -} -createRoot(document.getElementById('root')).render( - {mode === 'editor' ? : } -) -` - -const bundles = mobileWebAppDependenciesPresent() -const describeLayout = bundles ? describe : describe.skip -const browsers = new Map() -let scratch -let server -let origin - -beforeAll(async () => { - if (!bundles) { - return - } - await mkdir(join(mobileDir, '.tmp'), { recursive: true }) - scratch = await mkdtemp(join(mobileDir, '.tmp', 'queued-messages-')) - const outDir = join(scratch, 'bundle') - await esbuild.build({ - ...mobileWebAppBuildOptions([]), - entryPoints: undefined, - stdin: { contents: entry, resolveDir: mobileDir, loader: 'tsx', sourcefile: 'queued.tsx' }, - splitting: false, - write: true, - outdir: outDir, - entryNames: 'queued', - metafile: false - }) - await writeFile( - join(outDir, 'index.html'), - `${MOBILE_WEB_APP_ROOT_RESET}` + - `${MOBILE_WEB_APP_NATIVE_PARITY_STYLE}
` + - '' - ) - const served = await createBundleServer({ outDir, cspHeader: await readShellCsp() }) - server = served.server - origin = served.origin - for (const engine of engines) { - browsers.set(engine.name, await engine.launch()) - } -}, 180_000) - -afterAll(async () => { - for (const browser of browsers.values()) { - await browser.close() - } - server?.close() - if (scratch) { - await rm(scratch, { recursive: true, force: true }) - } -}) - -async function openPage(engine, hash, width) { - const page = await browsers.get(engine.name).newPage({ viewport: { width, height: 700 } }) - await page.goto(`${origin}/#${hash}`, { waitUntil: 'load' }) - return page -} - -function editorActionProblems(page) { - return page.evaluate(() => { - const row = document.querySelector('[data-testid="native-chat-composer-actions"]') - const problems = [] - const [cancel, save] = ['cancel', 'save'].map((name) => { - const button = row.querySelector(`[aria-label="${window.actionLabels[name]}"]`) - const label = button.querySelector('div[dir]') - const range = document.createRange() - range.selectNodeContents(label) - const text = range.getBoundingClientRect() - const box = button.getBoundingClientRect() - if (label.scrollWidth > label.clientWidth) { - problems.push(`${name} label truncated`) - } - if (text.left < box.left - 0.5 || text.right > box.right + 0.5) { - problems.push(`${name} label outside its button`) - } - return box - }) - if (cancel.right > save.left + 0.5) { - problems.push('cancel overlaps save') - } - if (save.right > row.getBoundingClientRect().right + 0.5) { - problems.push('save past the row') - } - return problems - }) -} - -describeLayout.each(engines)('mobile queued messages in $name', (engine) => { - it.each(['plain', 'drag'])('scrolls only as far as the %s rows reach', async (mode) => { - const page = await openPage(engine, mode, 390) - try { - await page.locator('[data-testid="queued-card-row"]').nth(1).waitFor() - const range = () => - page.evaluate(() => { - const box = document.querySelector('[data-testid="queued-box"]') - const rows = [...box.querySelectorAll('[data-testid="queued-card-row"]')] - const scroller = [...box.querySelectorAll('*')].find( - (node) => getComputedStyle(node).overflowY === 'auto' - ) - const rowsHeight = rows.reduce((sum, row) => sum + row.getBoundingClientRect().height, 0) - return scroller.scrollHeight - rowsHeight - }) - await expect.poll(range).toBeLessThanOrEqual(1) - } finally { - await page.close() - } - }) - - it('fits whole Save and Cancel labels at 360pt in every shipped locale', async () => { - const results = await Promise.all( - SUPPORTED_UI_LOCALES.map(async (locale) => { - const page = await openPage(engine, `editor:${locale}`, 360) - try { - await page.locator('[data-testid="native-chat-composer-actions"]').waitFor() - return { locale, problems: await editorActionProblems(page) } - } finally { - await page.close() - } - }) - ) - expect(results.filter(({ problems }) => problems.length > 0)).toEqual([]) - }, 60_000) -}) diff --git a/config/scripts/mobile-web-app/shell/mobile-web-app-render.test.mjs b/config/scripts/mobile-web-app/shell/mobile-web-app-render.test.mjs index 64cb4e18eb0..53848e466c1 100644 --- a/config/scripts/mobile-web-app/shell/mobile-web-app-render.test.mjs +++ b/config/scripts/mobile-web-app/shell/mobile-web-app-render.test.mjs @@ -17,7 +17,6 @@ import { // Mount real ExpoRoot: only rendering proves synthesized RequireContext has expected shape. const HOST_ROUTE = '/h/render-check-host' -const HOST_ROUTE_PATTERN = '/h/[hostId]' const SHELL_SESSION_ID = 'render-check-session' const SHELL_BUILD_ID = 'render-check-build' @@ -191,30 +190,6 @@ async function waitForRoute({ page, errors, uncaught }, route, awaitText) { } } -// Open / as shell does, then route from init; direct route navigation would hide entry routing bugs. -async function render(route, awaitText, { shellRoute = { pathname: route }, ...shell } = {}) { - const opened = await openPage({ shellRoute, ...shell }) - await opened.page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, route, awaitText) - const text = await opened.page.evaluate(() => document.body.innerText) - // Read page’s session identity to catch mounting without shell’s session or inventing one. - const session = await opened.page.evaluate(() => ({ - sessionId: document.documentElement.dataset.orcaWebSessionId ?? null, - buildId: document.documentElement.dataset.orcaWebBuildId ?? null - })) - // Require rewrite from / before render or every host route resolves to Unmatched. - const url = await opened.page.evaluate(() => location.pathname + location.search) - await opened.page.close() - // Name CSP refusals within asserted console errors for useful diagnostics. - return { - errors: opened.errors, - cspErrors: opened.errors.filter((entry) => entry.includes('Content Security Policy')), - text, - session, - url - } -} - const paintReports = (page, name) => page.evaluate( (paint) => @@ -223,58 +198,6 @@ const paintReports = (page, name) => ) describeRender('the Route A page in a real browser', () => { - it('fills the view, so what it mounted is painted and takes a tap', async () => { - const opened = await openPage({ shellRoute: { pathname: HOST_ROUTE } }) - await opened.page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, HOST_ROUTE, SHELL_HOST.name) - const layout = await opened.page.evaluate(() => { - // No-RPC control at root bottom moves furthest if layout collapses. - const fab = [...document.querySelectorAll('[role="button"]')].find( - (element) => element.getAttribute('aria-label') === 'New workspace' - ) - const box = fab?.getBoundingClientRect() ?? null - const hit = - box === null - ? null - : document.elementFromPoint(box.x + box.width / 2, box.y + box.height / 2) - return { - rootHeight: document.getElementById('root').getBoundingClientRect().height, - viewportHeight: window.innerHeight, - fabTop: box?.top ?? null, - fabBottom: box?.bottom ?? null, - reachesTheControl: hit !== null && fab.contains(hit) - } - }) - await opened.page.close() - expect(opened.errors).toEqual([]) - // Measure height: mounted text/innerText can pass while collapsed root paints blank phone screen. - expect(layout.rootHeight).toBe(layout.viewportHeight) - expect(layout.fabTop).toBeGreaterThan(0) - expect(layout.fabBottom).toBeLessThanOrEqual(layout.viewportHeight) - // Require reachable tap: clipped scroller rows retain layout rects. - expect(layout.reachesTheControl).toBe(true) - }, 60_000) - - // Native files routes throw requireNativeViewManager at import; deferred import hides this until actual route opens, requiring web siblings. - - it('mounts the file preview, reading the file path out of a param and not a segment', async () => { - const previewRoute = `${HOST_ROUTE}/files/preview/worktree-a` - const { errors, cspErrors, text, url } = await render(previewRoute, 'readme.md', { - shellRoute: { - pathname: previewRoute, - params: { relativePath: 'docs/my notes/readme.md', source: 'worktree' } - } - }) - expect(cspErrors).toEqual([]) - // Require no unsupported RN-Web BackHandler registration; Android hardware Back pops native stack while page Back owns unsaved-draft prompt. - expect(errors).toEqual([]) - // Title proves final path segment only; URL checks outbound encoding, not middle-segment decoding. - // Decode hazards are covered by mobile-file-path-route-encoding.test.ts and mobile-file-preview-route.test.ts. - expect(text).toContain('readme.md') - expect(url).toBe(`${previewRoute}?relativePath=docs%2Fmy+notes%2Freadme.md&source=worktree`) - expect(text).not.toContain(UNMATCHED) - }, 60_000) - it('tells the shell when a route chunk throws, rather than sitting on a blank page', async () => { const chunk = routeChunks['./h/[hostId]/index.tsx'] expect(chunk, Object.keys(routeChunks).join(' ')).toBeTruthy() @@ -345,128 +268,4 @@ describeRender('the Route A page in a real browser', () => { heldChunks.delete(path) } }, 120_000) - - it('says nothing for a redirect screen whose target is still behind its chunk', async () => { - // PR redirect remains mounted behind destination’s pending chunk; its own commit must not report screen paint. - const target = routeChunks['./h/[hostId]/source-control/[worktreeId].tsx'] - expect(target, Object.keys(routeChunks).join(' ')).toBeTruthy() - const path = `/assets/${target}` - let arrive = () => {} - heldChunks.set( - path, - new Promise((resolve) => { - arrive = resolve - }) - ) - try { - const route = `${HOST_ROUTE}/pr/render-check-tree` - const opened = await openPage({ - shellRoute: { pathname: route }, - shellPageRoutes: [HOST_ROUTE_PATTERN, '/h/[hostId]/pr/[worktreeId]'] - }) - await opened.page.goto(`${origin}/`, { waitUntil: 'load' }) - await opened.page.waitForFunction( - () => location.pathname.includes('/source-control/'), - undefined, - { timeout: 30_000, polling: 250 } - ) - // Router has moved but pending hub chunk still leaves document empty. - await waitForPageFrames(opened.page) - expect(await paintReports(opened.page, paintName)).toBe(0) - - arrive() - await opened.page.waitForFunction( - (name) => - (globalThis.__orcaRenderCheckNotifies ?? []).filter((frame) => frame.name === name) - .length > 0, - paintName, - { timeout: 30_000, polling: 250 } - ) - await opened.page.close() - } finally { - arrive() - heldChunks.delete(path) - } - }, 120_000) - - it('refuses a target the shell will not take, rather than opening it in the page', async () => { - // Fault-only grant makes notifyNavigate refuse like disposed/older shell; local fallback once mounted Unmatched or native-only session route. - const opened = await openPage({ shellRoute: { pathname: HOST_ROUTE } }) - const { page, errors } = opened - await page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, HOST_ROUTE, SHELL_HOST.name) - await page.getByLabel('Back to hosts').click() - // Settle handoff microtask before asserting no navigation. - await waitForPageFrames(page) - expect(await page.evaluate(() => location.pathname)).toBe(HOST_ROUTE) - const text = await page.evaluate(() => document.body.innerText) - expect(text).toContain(SHELL_HOST.name) - expect(text).not.toContain(UNMATCHED) - // Require no notify as well as unchanged page; successful shell handoff also leaves document in place. - const notifies = await page.evaluate(() => globalThis.__orcaRenderCheckNotifies ?? []) - expect(notifies.filter((frame) => frame.name === 'navigate')).toEqual([]) - // Refused target declines navigation without faulting page. - expect(await page.evaluate(() => globalThis.__orcaRenderCheckFaults ?? [])).toEqual([]) - expect(errors).toEqual([]) - await page.close() - }, 60_000) - - it("fetches the next route's chunks on a client-side navigation", async () => { - const opened = await openPage({ shellRoute: { pathname: HOST_ROUTE } }) - const { page, errors, scripts } = opened - await page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, HOST_ROUTE, SHELL_HOST.name) - const loadedForFirstRoute = [...scripts] - // Shell fetches document once; later route must arrive as a chunk through in-page history navigation. - await page.evaluate((to) => { - history.pushState(null, '', to) - dispatchEvent(new PopStateEvent('popstate')) - }, `${HOST_ROUTE}/tasks`) - await waitForRoute(opened, `${HOST_ROUTE}/tasks`, 'Issues') - expect(new URL(page.url()).pathname).toBe(`${HOST_ROUTE}/tasks`) - const fetchedOnNavigation = scripts.filter((path) => !loadedForFirstRoute.includes(path)) - // Require builder-identified tasks chunk, not any fetched script or code already loaded by entry. - const tasksChunk = routeChunks['./h/[hostId]/tasks.tsx'] - expect(tasksChunk, Object.keys(routeChunks).join(' ')).toBeTruthy() - expect(fetchedOnNavigation, scripts.join(' ')).toContain(`/assets/${tasksChunk}`) - expect(loadedForFirstRoute).not.toContain(`/assets/${tasksChunk}`) - const text = await page.evaluate(() => document.body.innerText) - expect(text).toContain('Tasks') - expect(text).not.toContain(UNMATCHED) - expect(errors).toEqual([]) - await page.close() - }, 60_000) -}) - -// Probe mounted React Navigation canGoBack through real route chevron/router.back; page exposes no direct test hook. -// First prove same-family tap reaches shell, since an inert handler and refused local Back both leave document unchanged. -describeRender('the stack the page Back button rests on', () => { - async function clickAndSettle(page, selector) { - await page.click(selector) - await waitForPageFrames(page) - return page.evaluate(() => location.pathname + location.search) - } - - it('carries a handoff the shell granted across the bridge from a real tap', async () => { - // Grant navigate: this chevron proves handler/bridge path used by navigate-back before interpreting absent notifies. - const opened = await openPage({ - shellRoute: { pathname: HOST_ROUTE }, - shellGrants: [faultGrant, 'navigate'], - shellPageRoutes: [HOST_ROUTE_PATTERN] - }) - await opened.page.goto(`${origin}/`, { waitUntil: 'load' }) - await waitForRoute(opened, HOST_ROUTE, SHELL_HOST.name) - const url = await clickAndSettle(opened.page, '[aria-label="Back to hosts"]') - await expect - .poll(() => - opened.page.evaluate(() => - (globalThis.__orcaRenderCheckNotifies ?? []).filter((frame) => frame.name === 'navigate') - ) - ) - .toEqual([{ v: bridgeVersion, type: 'notify', name: 'navigate', href: '/' }]) - // Require page stays on served route; local fallthrough would mount a screen document cannot serve. - expect(url).toBe(HOST_ROUTE) - expect(opened.errors).toEqual([]) - await opened.page.close() - }, 60_000) }) diff --git a/config/scripts/mobile-web-app/terminal/mobile-web-app-terminal-live-input-render.test.mjs b/config/scripts/mobile-web-app/terminal/mobile-web-app-terminal-live-input-render.test.mjs index 50cbd4995d4..afde47a339f 100644 --- a/config/scripts/mobile-web-app/terminal/mobile-web-app-terminal-live-input-render.test.mjs +++ b/config/scripts/mobile-web-app/terminal/mobile-web-app-terminal-live-input-render.test.mjs @@ -130,35 +130,6 @@ const fieldValue = (page) => describeRender( 'the terminal live input on the page', () => { - it('clears the field when Enter submits, the way it does natively', async () => { - // Use real keydown to exercise RN-Web Enter/onSubmitEditing; command must clear previously sent live text. - const { errors, page } = await openProbe() - await page.focus(`#${LIVE_INPUT_FIELD_ID}`) - await page.keyboard.type('ls') - await page.waitForFunction( - (id) => document.getElementById(id)?.value === 'ls', - LIVE_INPUT_FIELD_ID, - { timeout: 30_000, polling: 100 } - ) - - await page.keyboard.press('Enter') - - await page.waitForFunction( - () => (globalThis.__orcaLiveInputProbe.sent() ?? []).includes('\r'), - undefined, - { timeout: 30_000, polling: 100 } - ) - // Require exact output once: RN-Web cancels submitted keydown, so page line-break binding must not send second CR. - expect(await page.evaluate(() => globalThis.__orcaLiveInputProbe.sent())).toEqual([ - 'l', - 's', - '\r' - ]) - expect(await fieldValue(page)).toBe('') - expect(errors).toEqual([]) - await page.close() - }, 300_000) - it('submits when Enter arrives with the keyboard composition still open', async () => { // Android soft-keyboard Enter carries isComposing; RN-Web suppresses onSubmitEditing unlike native editor action. const { errors, page } = await openProbe() @@ -270,31 +241,6 @@ describeRender( expect(errors).toEqual([]) await page.close() }, 300_000) - - it('erases and retypes a word the keyboard corrects when it commits', async () => { - const { errors, page } = await openProbe({ userAgent: ANDROID_WEBVIEW_USER_AGENT }) - await page.focus(`#${LIVE_INPUT_FIELD_ID}`) - const input = await page.context().newCDPSession(page) - for (const text of ['t', 'te', 'teh']) { - await compose(input, text) - await waitForField(page, text) - } - - await input.send('Input.insertText', { text: 'the' }) - await waitForField(page, 'the') - - await page.waitForFunction( - () => globalThis.__orcaLiveInputProbe.sent().length === 4, - undefined, - { - timeout: 30_000, - polling: 50 - } - ) - expect(await sent(page)).toEqual(['t', 'e', 'h', '\u007f\u007fhe']) - expect(errors).toEqual([]) - await page.close() - }, 300_000) }) }, 900_000 diff --git a/config/scripts/performance/measurement/benchmark-sample-summary.test.mjs b/config/scripts/performance/measurement/benchmark-sample-summary.test.mjs deleted file mode 100644 index 2a1ac51b098..00000000000 --- a/config/scripts/performance/measurement/benchmark-sample-summary.test.mjs +++ /dev/null @@ -1,9 +0,0 @@ -import { describe, expect, it } from 'vitest' - -import { summarizeBenchmarkSamples } from './benchmark-sample-summary.mjs' - -describe('benchmark sample summary', () => { - it('averages the two middle samples for an even-sized median', () => { - expect(summarizeBenchmarkSamples([100, 1, 2, 99]).medianMs).toBe(50.5) - }) -}) diff --git a/config/scripts/static-analysis/process-boundaries/check-process-host-imports.test.mjs b/config/scripts/static-analysis/process-boundaries/check-process-host-imports.test.mjs deleted file mode 100644 index 6c5cf811520..00000000000 --- a/config/scripts/static-analysis/process-boundaries/check-process-host-imports.test.mjs +++ /dev/null @@ -1,48 +0,0 @@ -import { describe, expect, it } from 'vitest' -import { assessProcessHostImports } from './check-process-host-imports.mjs' - -const file = 'src/main/process-consumer.ts' -const manifest = { exports: { '.': './dist/index.js' } } -const assess = (source) => assessProcessHostImports(new Map([[file, source]]), manifest, new Set()) - -describe('process-host runtime import boundary', () => { - it.each([ - "import type { ChildProcess } from 'node:child_process'", - "import type * as ChildProcess from 'child_process'", - "import type ChildProcess = require('node:child_process')", - "export type { ChildProcess } from 'node:child_process'", - "type Child = import('node:child_process').ChildProcess" - ])('allows erased Node type ownership: %s', (source) => { - expect(assess(source)).toMatchObject({ direct: [], added: [], violations: [] }) - }) - - it.each([ - "import { spawn } from 'node:child_process'", - "import 'child_process'", - "import { type ChildProcess } from 'node:child_process'", - "export { spawn } from 'node:child_process'", - "const child = require('node:child_process')", - "const child = import('node:child_process')", - "import type { ChildProcess } from 'node:child_process'; require('node:child_process')", - "import { createRequire } from 'node:module'; const load = createRequire(import.meta.url); load('node:child_process')", - "process.getBuiltinModule('child_process')" - ])('still rejects runtime access: %s', (source) => { - expect(assess(source)).toMatchObject({ direct: [file], added: [file] }) - }) - - it('keeps package export restrictions for type-only imports', () => { - expect(assess("import type { Private } from '@orca/process-host/private'").violations).toEqual([ - `${file}: @orca/process-host/private is not a public process-host export` - ]) - }) - - it('requires the WSL allowlist only for runtime child-process imports', () => { - const wsl = "const program = 'wsl.exe'" - expect( - assess(`import type { ChildProcess } from 'node:child_process'; ${wsl}`).violations - ).toEqual([]) - expect(assess(`import { spawn } from 'node:child_process'; ${wsl}`).violations).toEqual([ - `${file}: direct wsl.exe exec must call assertWslDistroAllowed` - ]) - }) -}) diff --git a/config/scripts/test-runtime/environment/vitest-shared-worker-file-reset.test.mjs b/config/scripts/test-runtime/environment/vitest-shared-worker-file-reset.test.mjs deleted file mode 100644 index 4d10389b5b8..00000000000 --- a/config/scripts/test-runtime/environment/vitest-shared-worker-file-reset.test.mjs +++ /dev/null @@ -1,86 +0,0 @@ -import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs' -import { join, resolve } from 'node:path' -import { fileURLToPath } from 'node:url' -import { runProcess, runProcessSync } from '@orca/process-host' -import { expect, it } from 'vitest' -import { resolvePinnedBunExecutable } from '../../tool-invocation/pinned-bun-executable.mjs' - -const root = fileURLToPath(new URL('../../../..', import.meta.url)) - -it('restores environment stubs and finishes background imports before the next shared file', async () => { - const cache = join(root, 'node_modules/.cache') - mkdirSync(cache, { recursive: true }) - const directory = mkdtempSync(join(cache, 'shared-worker-reset-')) - const write = (name, source) => writeFileSync(join(directory, name), source) - try { - write( - 'vitest.config.mjs', - `import { defineConfig } from 'vitest/config' - import { BaseSequencer } from 'vitest/node' - class Ordered extends BaseSequencer { - async sort(specs) { return specs.sort((a, b) => a.moduleId.localeCompare(b.moduleId)) } - } - export default defineConfig({ test: { - root: ${JSON.stringify(directory)}, - include: ['*.spec.mjs'], isolate: false, maxWorkers: 1, - setupFiles: ['./assert-import-finished.mjs', ${JSON.stringify(resolve(root, 'config/scripts/test-runtime/environment/vitest-shared-worker-file-reset.mjs'))}], - sequence: { sequencer: Ordered } - } })` - ) - write( - 'assert-import-finished.mjs', - `import { afterAll, expect } from 'vitest' - import { readFileSync } from 'node:fs' - afterAll(() => expect(readFileSync(new URL('./result', import.meta.url), 'utf8')).toBe('loaded'))` - ) - write( - 'import-started.mjs', - `export let started - export const ready = new Promise(resolve => { started = resolve })` - ) - write( - 'background.mjs', - `import { setImmediate } from 'node:timers/promises' - import { writeFileSync } from 'node:fs' - import { started } from './import-started.mjs' - started() - await setImmediate() - writeFileSync(new URL('./result', import.meta.url), 'loaded')` - ) - write( - '01-replace-env.spec.mjs', - `import { it } from 'vitest' - import { ready } from './import-started.mjs' - it('replaces env and starts a background import', async () => { - process.env = { ...process.env } - void import('./background.mjs') - await ready - })` - ) - write( - '02-stub-env.spec.mjs', - `import { expect, it, vi } from 'vitest' - it('removes an environment override after a previous file replaced env', () => { - vi.stubEnv('ORCA_SHARED_WORKER_OVERRIDE', 'asia-east2') - vi.unstubAllEnvs() - expect(process.env.ORCA_SHARED_WORKER_OVERRIDE).toBeUndefined() - })` - ) - const result = await runProcess({ - program: resolvePinnedBunExecutable(root, runProcessSync) ?? 'bun', - args: [ - '--bun', - resolve(root, 'node_modules/vitest/vitest.mjs'), - 'run', - '--config', - join(directory, 'vitest.config.mjs') - ], - cwd: root, - env: { ...process.env, ORCA_BACKGROUND_LAUNCH: '1' }, - timeoutMs: 30_000 - }) - expect(result.code, result.stdout + result.stderr).toBe(0) - } finally { - rmSync(directory, { recursive: true, force: true }) - } -}) diff --git a/config/scripts/test-runtime/node-worker/vitest-node-runtime-pool.test.ts b/config/scripts/test-runtime/node-worker/vitest-node-runtime-pool.test.ts index 1d1d44efe69..9ba1a06ef16 100644 --- a/config/scripts/test-runtime/node-worker/vitest-node-runtime-pool.test.ts +++ b/config/scripts/test-runtime/node-worker/vitest-node-runtime-pool.test.ts @@ -1,95 +1,12 @@ -import { once } from 'node:events' import { mkdtemp, rm, writeFile } from 'node:fs/promises' import { tmpdir } from 'node:os' -import { join, resolve } from 'node:path' +import { join } from 'node:path' import { PassThrough } from 'node:stream' -import { afterAll, afterEach, beforeAll, expect, it, onTestFinished, vi } from 'vitest' -import { createVitest, type PoolOptions, type Vitest } from 'vitest/node' -import * as processes from '@orca/process-host' -import type { ChildProcess } from 'node:child_process' +import { expect, it, onTestFinished } from 'vitest' +import { createVitest } from 'vitest/node' + import { nodeRuntimePool } from './vitest-node-runtime-pool' -let context: Vitest -const children: ChildProcess[] = [] -const spawn = processes.spawnProcess - -beforeAll(async () => { - context = await createVitest('test', { config: false, watch: false, include: [] }) -}) - -afterEach(async () => { - vi.useRealTimers() - vi.restoreAllMocks() - for (const child of children.splice(0)) { - if (child.pid && child.exitCode === null && child.signalCode === null) { - const exited = once(child, 'exit') - child.kill('SIGKILL') - await exited - } - } -}) - -afterAll(async () => { - await context.close() -}) - -function options(executable = process.execPath): PoolOptions { - return { - distPath: '', - project: context.getRootProject(), - method: 'run', - environment: { name: 'node', options: null }, - execArgv: [], - env: { ...process.env, ORCA_TEST_NODE_EXECUTABLE: executable } - } -} - -function captureChild(): void { - vi.spyOn(processes, 'spawnProcess').mockImplementation((spec) => { - const child = spawn(spec) - children.push(child) - return child - }) -} - -it('rejects an unavailable Node executable without an uncaught error or leaked pipes', async () => { - const output = context.logger.outputStream - const before = output.getMaxListeners() - const worker = nodeRuntimePool.createPoolWorker(options(resolve('missing-node-executable'))) - await expect(worker.start()).rejects.toMatchObject({ code: 'ENOENT' }) - await worker.stop() - expect(output.getMaxListeners()).toBe(before) -}) - -it('waits for a running Node worker to exit and restores its output listener budget', async () => { - captureChild() - const output = context.logger.outputStream - const before = output.getMaxListeners() - const worker = nodeRuntimePool.createPoolWorker(options()) - await worker.start() - await worker.stop() - expect(children[0].signalCode).toBe('SIGTERM') - expect(output.getMaxListeners()).toBe(before) -}) - -it('reports a failed kill within its deadline rather than hanging teardown', async () => { - captureChild() - const worker = nodeRuntimePool.createPoolWorker(options()) - const output = context.logger.outputStream - const before = output.getMaxListeners() - await worker.start() - const child = children[0] - vi.spyOn(child, 'kill').mockReturnValue(false) - vi.useFakeTimers() - const stopped = expect(worker.stop()).rejects.toThrow('did not exit after SIGKILL') - await vi.advanceTimersByTimeAsync(2_000) - await stopped - expect(child.kill).toHaveBeenCalledWith('SIGTERM') - expect(child.kill).toHaveBeenCalledWith('SIGKILL') - expect(child.listenerCount('exit')).toBe(0) - expect(output.getMaxListeners()).toBe(before) -}) - it('reports a test file that fails to load as a failed file instead of crashing the run', async () => { const dir = await mkdtemp(join(tmpdir(), 'orca-node-runtime-pool-')) onTestFinished(() => rm(dir, { recursive: true, force: true })) diff --git a/config/scripts/test-runtime/sqlite-admission/vitest-node-runtime-files.mjs b/config/scripts/test-runtime/sqlite-admission/vitest-node-runtime-files.mjs index e4ff67a29f4..23a24017a42 100644 --- a/config/scripts/test-runtime/sqlite-admission/vitest-node-runtime-files.mjs +++ b/config/scripts/test-runtime/sqlite-admission/vitest-node-runtime-files.mjs @@ -132,7 +132,6 @@ const NODE_RUNTIME_INCLUDE = [ 'src/shared/ephemeral-vm-recipe-output-capture.test.ts', 'src/main/window/main-window-updater-deferred-setup.test.ts', 'src/renderer/src/components/terminal-pane/host-state/terminal-pane-host-state-memo.test.ts', - 'src/renderer/src/components/automations/runs/use-automation-runs-dashboard-payload.test.tsx', 'src/main/native-chat/agent-session-journal/**/*.test.ts', 'src/main/native-chat/agent-session-wire/host/testing/structured-agent-session-terminal-writer-acquire.test.ts', 'src/main/runtime/agent-session/records/agent-session-conversation-name-store.test.ts', diff --git a/config/scripts/workflow/source-routing/pr-e2e-source-routing.mjs b/config/scripts/workflow/source-routing/pr-e2e-source-routing.mjs index 9563c85d9a6..ff23fadcffa 100644 --- a/config/scripts/workflow/source-routing/pr-e2e-source-routing.mjs +++ b/config/scripts/workflow/source-routing/pr-e2e-source-routing.mjs @@ -153,16 +153,6 @@ export const PR_E2E_SOURCE_ROUTES = [ file ) }, - { - id: 'ssh.orcad-auto-convert', - specs: ['tests/e2e/ssh-orcad-auto-convert.spec.ts'], - matches: (file) => - MANAGED_SSH_HARNESS.test(file) || - (isProductSource(file) && - /^src\/main\/(?:ipc\/(?:ssh\/)?ssh-host-server-|(?:ssh|managed-server\/[a-z-]+)\/(?:ssh-host-server-|orcad-runtime-conversion|orcad-migration-|orcad-retained-source|orcad-runtime-deployment))/.test( - file - )) - }, { id: 'ssh.orcad-connect-failure', specs: [ diff --git a/mobile/src/session/native-chat/agent-session/mobile-structured-agent-session-launch.test.ts b/mobile/src/session/native-chat/agent-session/mobile-structured-agent-session-launch.test.ts deleted file mode 100644 index 13c056860b2..00000000000 --- a/mobile/src/session/native-chat/agent-session/mobile-structured-agent-session-launch.test.ts +++ /dev/null @@ -1,87 +0,0 @@ -import { describe, expect, it, vi } from 'vitest' -import type { RpcClient } from '../../../transport/rpc/rpc-client-contract' -import { markRpcDeliveryUnknown } from '../../../transport/rpc/rpc-delivery-ambiguity' -import { createMobileStructuredAgentSession } from './mobile-structured-agent-session-launch' -import './mobile-created-structured-sessions' - -function clientReturning( - ...responses: unknown[] -): RpcClient & { sendRequest: ReturnType } { - let responseIndex = 0 - const sendRequest = vi.fn(async () => responses[responseIndex++]) - return { sendRequest } as unknown as RpcClient & { sendRequest: ReturnType } -} - -describe('mobile structured agent-session launch', () => { - it('keeps the outcome unknown when the idempotent retry cannot be sent', async () => { - const client = clientReturning({ ok: true, result: { supported: true } }) - client.sendRequest.mockImplementationOnce(async () => ({ - ok: true, - result: { supported: true } - })) - client.sendRequest.mockRejectedValueOnce(markRpcDeliveryUnknown(new Error('response lost'))) - client.sendRequest.mockRejectedValueOnce(new Error('connection interrupted')) - - await expect( - createMobileStructuredAgentSession(client, 'workspace-1', 'codex') - ).resolves.toMatchObject({ - kind: 'unknown' - }) - }) -}) - -it('settles a proven missing-original fork refusal but keeps a lost earlier attempt unknown', async () => { - const refusal = { - ok: true, - result: { - ok: false, - refusal: { - code: 'agent_session_identity_required', - details: { reason: 'recordMissing' }, - createAttemptRefusedBeforeCommit: true, - message: 'Source chat unavailable' - } - } - } - const client = clientReturning({ ok: true, result: { supported: true } }, refusal) - const forkFrom = { sessionId: 'copied-chat', throughItemId: 'legacy:opencode:source:reply-1' } - expect( - (await createMobileStructuredAgentSession(client, 'workspace-1', 'opencode', forkFrom)).kind - ).toBe('failed') - const lost = clientReturning({ ok: true, result: { supported: true } }, refusal) - lost.sendRequest - .mockImplementationOnce(async () => ({ ok: true, result: { supported: true } })) - .mockRejectedValueOnce(markRpcDeliveryUnknown(new Error('reply lost'))) - .mockResolvedValueOnce(refusal) - expect( - (await createMobileStructuredAgentSession(lost, 'workspace-1', 'opencode', forkFrom)).kind - ).toBe('unknown') -}) - -it('settles a proven fork read refusal without settling a prior lost copy', async () => { - const refusal = { - ok: true, - result: { - ok: false, - refusal: { - code: 'agent_session_identity_required', - details: { reason: 'transcriptUnreadable' }, - createAttemptRefusedBeforeCommit: true, - message: 'Codex could not read the saved chat.' - } - } - } - const client = clientReturning({ ok: true, result: { supported: true } }, refusal) - const forkFrom = { sessionId: 'source-chat', throughItemId: 'codex:source:selected:1' } - expect( - (await createMobileStructuredAgentSession(client, 'workspace-1', 'codex', forkFrom)).kind - ).toBe('failed') - const lost = clientReturning({ ok: true, result: { supported: true } }, refusal) - lost.sendRequest - .mockImplementationOnce(async () => ({ ok: true, result: { supported: true } })) - .mockRejectedValueOnce(markRpcDeliveryUnknown(new Error('copy reply lost'))) - .mockResolvedValueOnce(refusal) - expect( - (await createMobileStructuredAgentSession(lost, 'workspace-1', 'codex', forkFrom)).kind - ).toBe('unknown') -}) diff --git a/mobile/src/session/native-chat/agent-session/use-mobile-structured-agent-session.test.tsx b/mobile/src/session/native-chat/agent-session/use-mobile-structured-agent-session.test.tsx deleted file mode 100644 index 73c4f8898d9..00000000000 --- a/mobile/src/session/native-chat/agent-session/use-mobile-structured-agent-session.test.tsx +++ /dev/null @@ -1,223 +0,0 @@ -import { createElement } from 'react' -import { - snapshotEvent as publishedSnapshot, - fieldsOf, - mutationOk, - ok -} from './use-mobile-structured-agent-session-queued.test-fixture' -import { act, type ReactTestRenderer } from 'react-test-renderer' -import { createWithQueryClient } from '../../../test-support/query-client-test-renderer' -import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' -import type { AgentJournalRenderItem } from '../../../../../src/shared/agent-session/journal/agent-session-journal-types' - -import type { RpcClient } from '../../../transport/rpc/rpc-client-contract' -import type { ConnectionState } from '../../../transport/connection/connection-state' - -import { structuredSendResultFixture } from '../../structured-agent-send-result.test-fixture' -import { useMobileStructuredAgentSession } from './use-mobile-structured-agent-session' - -const asyncStorage = vi.hoisted(() => ({ - getItem: vi.fn(), - setItem: vi.fn(), - removeItem: vi.fn() -})) - -vi.mock('@react-native-async-storage/async-storage', () => ({ default: asyncStorage })) - -function snapshotEvent(fence = 3, items: AgentJournalRenderItem[] = []) { - return publishedSnapshot({ fence, items }) -} - -function snapshotWithMessage() { - return snapshotEvent(3, [ - { - itemId: 'msg-1', - revision: 1, - sequence: 1, - observedAt: 10, - body: { - kind: 'message', - role: 'user', - blocks: [{ type: 'text', text: 'sent before the blip' }] - } - } - ]) -} - -async function defaultSendRequest(method: string, params?: unknown) { - if (method === 'agentSession.send') { - return mutationOk(structuredSendResultFixture('accepted')) - } - if (method === 'agentSession.options') { - return ok({ - models: [ - { - id: 'gpt-fast', - label: 'GPT Fast', - isDefault: true, - defaultEffort: 'low', - efforts: [ - { value: 'low', label: 'Low' }, - { value: 'high', label: 'High' } - ] - }, - { - id: 'gpt-slow', - label: 'GPT Slow', - isDefault: false, - defaultEffort: 'high', - efforts: [ - { value: 'low', label: 'Low' }, - { value: 'high', label: 'High' } - ] - } - ], - current: { - model: 'gpt-fast', - effort: 'low' - } - }) - } - if (method === 'agentSession.setOption') { - return ok({ - ok: true, - replayed: false, - fence: 3, - cursor: { epoch: 'epoch-1', sequence: 2 }, - value: { - key: 'model', - value: 'gpt-fast', - options: { model: 'gpt-fast' } - } - }) - } - if (method === 'agentSession.respondToApproval' || method === 'agentSession.respondToQuestion') { - return ok({ - ok: true, - replayed: false, - fence: 3, - cursor: { epoch: 'epoch-1', sequence: 3 }, - value: { - itemId: String(fieldsOf(params).itemId ?? ''), - revision: 2, - resolution: { - state: 'resolved', - selectedOptionId: String(fieldsOf(params).optionId ?? ''), - resolvedBy: 'mobile', - resolvedAt: 123 - } - } - }) - } - return ok({}) -} - -describe('useMobileStructuredAgentSession', () => { - let renderer: ReactTestRenderer | null = null - let hook: ReturnType | null = null - let listener: ((value: unknown) => void) | null = null - const onSendError = vi.fn() - const unsubscribe = vi.fn() - const unsubscribePreference = vi.fn() - const sendRequest = vi.fn(defaultSendRequest) - const subscribe = vi.fn((_method: string, _params: unknown, onData: (value: unknown) => void) => { - if (_method === 'agentSession.subscribe') { - listener = onData - return unsubscribe - } - return unsubscribePreference - }) - let clientState: ConnectionState = 'connected' - const stateListeners = new Set<(state: ConnectionState) => void>() - - const client: RpcClient = { - sendRequest, - subscribe, - getState: () => clientState, - onStateChange: (stateListener) => { - stateListeners.add(stateListener) - return () => stateListeners.delete(stateListener) - }, - getReconnectAttempt: () => 0, - getLastConnectedAt: () => null, - updateTerminalSubscriptionViewport: () => {}, - notifyForeground: () => {}, - close: () => {} - } - let storedOperations: Map - - function Harness({ - sessionId = 'session-1', - agent = 'codex', - connected = true, - sourceIdentity = 'host-a\0workspace-a' - }: { - sessionId?: string | null - agent?: string | null - connected?: boolean - sourceIdentity?: string - }): null { - hook = useMobileStructuredAgentSession({ - client, - pairedHostClient: client, - sessionId, - sourceIdentity, - enabled: true, - connected, - hostSupport: null, - agent, - onSendError - }) - return null - } - - beforeEach(() => { - vi.clearAllMocks() - - storedOperations = new Map() - asyncStorage.getItem.mockImplementation( - async (key: string) => storedOperations.get(key) ?? null - ) - asyncStorage.setItem.mockImplementation(async (key: string, value: string) => { - storedOperations.set(key, value) - }) - asyncStorage.removeItem.mockImplementation(async (key: string) => { - storedOperations.delete(key) - }) - sendRequest.mockImplementation(defaultSendRequest) - listener = null - clientState = 'connected' - }) - - afterEach(() => { - act(() => renderer?.unmount()) - renderer = null - hook = null - }) - - it('isolates matching provider session ids across host and workspace sources', async () => { - await act(async () => { - renderer = createWithQueryClient( - createElement(Harness, { - connected: true, - sessionId: 'session-1', - sourceIdentity: 'host-a\0workspace-a' - }) - ) - }) - await vi.waitFor(() => expect(listener).toEqual(expect.any(Function))) - act(() => listener?.(snapshotWithMessage())) - expect(hook?.session.messages).toHaveLength(1) - - await act(async () => { - renderer?.update( - createElement(Harness, { - connected: false, - sessionId: 'session-1', - sourceIdentity: 'host-b\0workspace-b' - }) - ) - }) - expect(hook?.session.messages).toEqual([]) - }) -}) diff --git a/mobile/src/source-control/folder/folder-repo-worktree-create.test.ts b/mobile/src/source-control/folder/folder-repo-worktree-create.test.ts deleted file mode 100644 index e89ca544001..00000000000 --- a/mobile/src/source-control/folder/folder-repo-worktree-create.test.ts +++ /dev/null @@ -1,332 +0,0 @@ -import { afterEach, describe, expect, it, vi } from 'vitest' -import { FakeSession } from '../../transport/connection/mobile-endpoint-supervisor-test-fakes' -import { markRpcDeliveryUnknown } from '../../transport/rpc/rpc-delivery-ambiguity' -import type { RpcResponse } from '../../transport/rpc/rpc-wire-types' -import { - beginFolderComposerSubmission, - createFolderRepoWorktree, - isFolderRepoCreateBusy, - reconcileFolderRepoCreates, - releaseFolderComposerReservation, - type FolderRepoTarget -} from './folder-repo-worktree-create' - -const idempotency = vi.hoisted(() => { - const state: { support: { dedupeTtlMs: number } | false } = { support: false } - return state -}) -vi.mock('../../tasks/worktree-create-capability', () => ({ - readNewWorktreeRuntimeCapabilities: async () => ({ - worktreeCreateIdempotency: idempotency.support - }) -})) - -const FOLDER = 'folder:f-1' -const TAKEN = 'Branch payments already exists\nworktree_create_exact_name_taken' - -afterEach(() => { - vi.useRealTimers() - idempotency.support = false -}) - -let repoCount = 0 -// Each test's own repository: the registry is shared, as it is across screens. -function target(executionHost?: string): FolderRepoTarget { - repoCount += 1 - return { hostId: 'host-1', executionHost, folderWorktreeId: FOLDER, repoId: `repo-${repoCount}` } -} - -function refused(message: string): RpcResponse { - return { id: '1', ok: false, error: { code: 'x', message }, _meta: { runtimeId: 'r' } } -} - -const CREATED: RpcResponse = { - id: '1', - ok: true, - result: { worktree: { id: 'repo::/w/payments', displayName: 'payments' } }, - _meta: { runtimeId: 'r' } -} - -function host(...replies: (RpcResponse | Error)[]): FakeSession { - const client = new FakeSession('connected') - for (const reply of replies) { - if (reply instanceof Error) { - client.sendRequest.mockRejectedValueOnce(reply) - } else { - client.sendRequest.mockResolvedValueOnce(reply) - } - } - return client -} - -const create = (repo: FolderRepoTarget, client: FakeSession) => - createFolderRepoWorktree({ target: repo, client, name: 'payments' }) - -describe('createFolderRepoWorktree', () => { - it('sends one exact-name create attached to the folder', async () => { - const repo = target() - const client = host(CREATED) - - await expect(create(repo, client)).resolves.toEqual({ kind: 'created', name: 'payments' }) - expect(client.sendRequest.mock.calls).toEqual([ - [ - 'worktree.create', - { - repo: `id:${repo.repoId}`, - name: 'payments', - displayName: 'payments', - displayNameKind: 'user', - branchIntent: 'create-exact', - setupDecision: 'inherit', - parentWorkspace: FOLDER, - parentWorkspaceOrigin: 'manual' - }, - expect.anything() - ] - ]) - }) - - it('sends nothing for a second tap while the first create runs', async () => { - const repo = target() - const client = new FakeSession('connected') - let answer: (response: RpcResponse) => void = () => {} - client.sendRequest.mockReturnValue(new Promise((resolve) => (answer = resolve))) - - const first = create(repo, client) - await expect(create(repo, client)).resolves.toEqual({ kind: 'already-running' }) - answer(CREATED) - await first - expect(client.sendRequest).toHaveBeenCalledTimes(1) - }) - - it.each([ - ['the name is taken', TAKEN], - ['the repo asks about setup', 'Setup decision required for this repository'] - ])('keeps the repository for the composer when %s', async (_case, message) => { - const repo = target() - const outcome = await create(repo, host(refused(message))) - - expect(outcome.kind).toBe('needs-composer') - expect(isFolderRepoCreateBusy(repo)).toBe(true) - if (outcome.kind === 'needs-composer') { - releaseFolderComposerReservation(repo, outcome.reservation) - } - expect(isFolderRepoCreateBusy(repo)).toBe(false) - }) - - it('treats a lost reply as possibly created, past the busy backstop, and never offers the composer', async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const repo = target() - const lost = markRpcDeliveryUnknown(new Error('connection lost')) - - await expect(create(repo, host(lost))).resolves.toEqual({ kind: 'maybe-created' }) - expect(isFolderRepoCreateBusy(repo)).toBe(true) - await expect(create(repo, host(CREATED))).resolves.toEqual({ kind: 'already-running' }) - - // The list never showed it; the card takes taps again but the create is still unresolved. - vi.advanceTimersByTime(60_000) - expect(isFolderRepoCreateBusy(repo)).toBe(false) - const retry = host(refused(TAKEN)) - await expect(create(repo, retry)).resolves.toEqual({ kind: 'maybe-created' }) - expect(retry.sendRequest).toHaveBeenCalledTimes(1) - }) - - it('retries an unconfirmed create as the name it was sent as, even after the folder is renamed', async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const repo = target() - await create(repo, host(markRpcDeliveryUnknown(new Error('connection lost')))) - vi.advanceTimersByTime(60_000) - - const retry = host(refused(TAKEN)) - await createFolderRepoWorktree({ target: repo, client: retry, name: 'billing' }) - expect(retry.sendRequest.mock.calls[0]?.[1]).toMatchObject({ name: 'payments' }) - }) - - it('asks for setup when an unconfirmed create is refused for it, and resends the same exact name', async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const repo = target() - await create(repo, host(markRpcDeliveryUnknown(new Error('connection lost')))) - vi.advanceTimersByTime(60_000) - - const asked = await create(repo, host(refused('Setup decision required for this repository'))) - expect(asked.kind).toBe('needs-setup-choice') - expect(isFolderRepoCreateBusy(repo)).toBe(false) - const hold = asked.kind === 'needs-setup-choice' ? asked.hold : -1 - - const resent = host(refused(TAKEN)) - await expect( - createFolderRepoWorktree({ - target: repo, - client: resent, - name: 'billing', - skipSetupFor: hold - }) - ).resolves.toEqual({ kind: 'maybe-created' }) - expect(resent.sendRequest.mock.calls[0]?.[1]).toMatchObject({ - name: 'payments', - branchIntent: 'create-exact', - setupDecision: 'skip' - }) - }) - - it('sends nothing for a setup choice once its create is listed or replaced', async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const ask = async (repo: FolderRepoTarget) => { - await create(repo, host(markRpcDeliveryUnknown(new Error('connection lost')))) - vi.advanceTimersByTime(60_000) - const asked = await create(repo, host(refused('Setup decision required for this repository'))) - return asked.kind === 'needs-setup-choice' ? asked.hold : -1 - } - const listed = target() - const listedHold = await ask(listed) - const folder = { hostId: 'host-1', executionHost: undefined, folderWorktreeId: FOLDER } - reconcileFolderRepoCreates(folder, [{ repoId: listed.repoId }]) - const replaced = target() - const replacedHold = await ask(replaced) - await ask(replaced) - - const late = host(CREATED) - for (const [repo, hold] of [ - [listed, listedHold], - [replaced, replacedHold] - ] as const) { - await expect( - createFolderRepoWorktree({ - target: repo, - client: late, - name: 'billing', - skipSetupFor: hold - }) - ).resolves.toEqual({ kind: 'checking' }) - } - expect(late.sendRequest).not.toHaveBeenCalled() - }) -}) - -describe('createFolderRepoWorktree after a lost reply on a host that dedupes', () => { - it.each([ - ['answers with the worktree', CREATED, { kind: 'created', name: 'payments' }], - ['refuses the name as taken', refused(TAKEN), { kind: 'maybe-created' }] - ])('resends the same create once reconnected; the host %s', async (_case, replay, expected) => { - idempotency.support = { dedupeTtlMs: 60_000 } - const repo = target() - const client = new FakeSession('connected') - client.sendRequest - .mockImplementationOnce(async () => { - client.publishState('reconnecting') - setTimeout(() => client.publishState('connected'), 0) - throw markRpcDeliveryUnknown(new Error('connection lost')) - }) - .mockResolvedValueOnce(replay) - - await expect(create(repo, client)).resolves.toEqual(expected) - const [first, second] = client.sendRequest.mock.calls.map(([, params]) => params) - expect(second).toEqual(first) - expect(first).toHaveProperty('clientMutationId') - }) -}) - -describe('reconcileFolderRepoCreates', () => { - it('forgets a create once the folder lists its worktree on the same host, not another', async () => { - const repo = target('runtime:env-1') - await create(repo, host(CREATED)) - const folder = { hostId: 'host-1', executionHost: 'runtime:env-1', folderWorktreeId: FOLDER } - - reconcileFolderRepoCreates(folder, [{ repoId: repo.repoId, hostId: 'runtime:env-2' }]) - expect(isFolderRepoCreateBusy(repo)).toBe(true) - reconcileFolderRepoCreates(folder, [{ repoId: repo.repoId, hostId: 'runtime:env-1' }]) - expect(isFolderRepoCreateBusy(repo)).toBe(false) - }) - - it('keeps a create still on its way even if the list already shows the repository', async () => { - const repo = target() - const client = new FakeSession('connected') - let answer: (response: RpcResponse) => void = () => {} - client.sendRequest.mockReturnValue(new Promise((resolve) => (answer = resolve))) - const folder = { hostId: 'host-1', executionHost: undefined, folderWorktreeId: FOLDER } - const members = [{ repoId: repo.repoId }] - - const pending = create(repo, client) - reconcileFolderRepoCreates(folder, members) - expect(isFolderRepoCreateBusy(repo)).toBe(true) - answer(CREATED) - await pending - reconcileFolderRepoCreates(folder, members) - expect(isFolderRepoCreateBusy(repo)).toBe(false) - }) - - it("doesn't let an earlier create's timer end a newer one", async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const repo = target() - await create(repo, host(markRpcDeliveryUnknown(new Error('lost')))) - vi.advanceTimersByTime(10_000) - const folder = { hostId: 'host-1', executionHost: undefined, folderWorktreeId: FOLDER } - reconcileFolderRepoCreates(folder, [{ repoId: repo.repoId }]) - await create(repo, host(CREATED)) - - vi.advanceTimersByTime(10_000) - expect(isFolderRepoCreateBusy(repo)).toBe(true) - }) -}) - -describe('beginFolderComposerSubmission', () => { - async function composing() { - const repo = target() - const outcome = await create(repo, host(refused(TAKEN))) - if (outcome.kind !== 'needs-composer') { - throw new Error('expected the composer') - } - return { repo, launch: { target: repo, reservation: outcome.reservation, name: 'payments' } } - } - - it('admits one create per repository while it is out, even after the form closes', async () => { - const { repo, launch } = await composing() - let formOpen = true - const submission = beginFolderComposerSubmission(repo, launch, () => formOpen) - - expect(submission).not.toBeNull() - formOpen = false - releaseFolderComposerReservation(repo, launch.reservation) - expect(beginFolderComposerSubmission(repo, launch, () => true)).toBeNull() - expect(await create(repo, host(CREATED))).toEqual({ kind: 'already-running' }) - - submission?.settle('uncertain') - expect(isFolderRepoCreateBusy(repo)).toBe(true) - }) - - it('gives a refused create back to the open form, and a same-host switch its own admission', async () => { - const { repo, launch } = await composing() - const other = { ...repo, repoId: `${repo.repoId}-other` } - - beginFolderComposerSubmission(repo, launch, () => true)?.settle('failed') - expect(isFolderRepoCreateBusy(repo)).toBe(true) - const switched = beginFolderComposerSubmission(other, launch, () => true) - expect(switched).not.toBeNull() - expect(beginFolderComposerSubmission(other, launch, () => true)).toBeNull() - switched?.settle('created') - releaseFolderComposerReservation(repo, launch.reservation) - expect(isFolderRepoCreateBusy(repo)).toBe(false) - expect(isFolderRepoCreateBusy(other)).toBe(true) - }) - it("never sends anything from a card for a form's unconfirmed create, whatever name it used", async () => { - vi.useFakeTimers({ toFake: ['setTimeout'] }) - const { repo, launch } = await composing() - // The form may have sent another name or branch than the card's. - beginFolderComposerSubmission(repo, launch, () => true)?.settle('uncertain') - vi.advanceTimersByTime(60_000) - - expect(beginFolderComposerSubmission(repo, launch, () => true)).toBeNull() - const tap = host(CREATED) - await expect(create(repo, tap)).resolves.toEqual({ kind: 'checking' }) - await expect( - createFolderRepoWorktree({ - target: repo, - client: tap, - name: 'payments', - skipSetupFor: 0 - }) - ).resolves.toEqual({ kind: 'checking' }) - expect(tap.sendRequest).not.toHaveBeenCalled() - expect(isFolderRepoCreateBusy(repo)).toBe(false) - }) -}) diff --git a/src/main/acp/discovery/acp-model-catalog-probe.test.ts b/src/main/acp/discovery/acp-model-catalog-probe.test.ts deleted file mode 100644 index 85f040a7fb3..00000000000 --- a/src/main/acp/discovery/acp-model-catalog-probe.test.ts +++ /dev/null @@ -1,196 +0,0 @@ -import { mkdtemp, rm } from 'node:fs/promises' -import { tmpdir } from 'node:os' -import { join } from 'node:path' -import { afterEach, describe, expect, it } from 'vitest' -import { acpLaunchSpecFor } from '../launch/acp-launch-registry' -import type { AcpLaunchSpec } from '../launch/acp-launch-contract' -import { createAcpModelCatalogProbe } from './acp-model-catalog-probe' -import { AcpSessionRuntime } from '../protocol/acp-session-runtime' -import { AcpScriptedAgent } from '../process/testing/acp-scripted-agent.test-support' -import { openCodeAcpAccountBinding } from '../../opencode/opencode-structured-account-home' -import { AcpAgentExitedError } from '../transport/acp-errors' -// Real `omp acp` 18.17.0 session/new replies: started on its default fake/wide, with -// `--model fake/narrow`, and with `--model fake/gone`, which it cannot resolve. -import recorded from '../fixtures/omp-v18.17.0-model-efforts.json' - -// Bypass the app's managed-account service for this fixture. -const OPENCODE: AcpLaunchSpec = { - ...acpLaunchSpecFor('opencode')!, - account: openCodeAcpAccountBinding(() => ({ - list: () => ({ activeAccountId: null, accounts: [] }), - restoreOriginalEnvironment: () => {}, - environmentForAccount: () => ({}) - })) -} - -const agents: AcpScriptedAgent[] = [] -afterEach(() => { - for (const agent of agents.splice(0)) { - agent.close() - } -}) - -describe('ACP model catalog probes', () => { - it('lists the catalog after OpenCode answers session/new with -32603 while starting', async () => { - const agent = new AcpScriptedAgent() - agents.push(agent) - agent.on('initialize', (frame) => - agent.reply(frame, { - protocolVersion: 1, - agentCapabilities: { sessionCapabilities: { delete: {} } } - }) - ) - let sessionNews = 0 - agent.on('session/new', (frame) => { - sessionNews += 1 - if (sessionNews === 1) { - agent.fail(frame, -32603, 'Internal error: Internal service failure', { - errorName: 'InvalidRequestError' - }) - return - } - agent.reply(frame, { - sessionId: 'probe-session', - configOptions: [ - { - id: 'model', - name: 'Model', - category: 'model', - type: 'select', - currentValue: 'opencode/exo-free', - options: [ - { value: 'opencode/big-pickle', name: 'Big Pickle' }, - { value: 'opencode/exo-free', name: 'Exo Free' } - ] - } - ] - }) - }) - agent.on('session/delete', (frame) => agent.reply(frame, {})) - agent.on('session/set_config_option', (frame) => agent.fail(frame, -32602, 'Model unavailable')) - - const success = await createAcpModelCatalogProbe(OPENCODE, { - resolveEnvironment: async () => ({ PATH: '/usr/bin', HOME: '/home/user' }), - resolveCommand: () => '/opt/opencode/bin/opencode', - probeVersion: async () => ({ version: '2.0.29' }), - homePath: '/home/user', - dataDirectory: '/orca-data', - connectSession: () => { - const runtime = new AcpSessionRuntime(agent.stdout, agent.stdin) - return { - initialize: () => runtime.initialize(), - start: (options) => runtime.start(options), - setConfigOption: (id, value) => runtime.setConfigOption(id, value), - deleteSession: (sessionId) => runtime.deleteSession(sessionId), - close: async () => runtime.close() - } - } - })({ kind: 'opencode', locator: { kind: 'unmanaged' } }, { workspacePath: '/work/project' }) - - expect(sessionNews).toBe(2) - expect(success.models.map((model) => model.id)).toEqual([ - 'opencode/big-pickle', - 'opencode/exo-free' - ]) - }) -}) - -async function listOmp(selectedModels?: string[]) { - const agent = new AcpScriptedAgent() - agents.push(agent) - let launchedModel: string | undefined - agent.on('initialize', (frame) => agent.reply(frame, { protocolVersion: 1 })) - agent.on('session/new', (frame) => - agent.reply(frame, { - sessionId: 'probe-session', - ...(launchedModel === 'fake/narrow' - ? recorded.narrow - : launchedModel === 'fake/gone' - ? recorded.unresolved - : recorded.new) - }) - ) - const dataDirectory = await mkdtemp(join(tmpdir(), 'orca-omp-efforts-')) - try { - const listing = await createAcpModelCatalogProbe(acpLaunchSpecFor('omp')!, { - resolveEnvironment: async () => ({ PATH: '/usr/bin', HOME: '/home/user' }), - resolveCommand: () => '/opt/omp/bin/omp', - probeVersion: async () => ({ version: '18.17.0' }), - homePath: '/home/user', - dataDirectory, - connectSession: (launch) => { - const at = launch.args.indexOf('--model') - launchedModel = at === -1 ? undefined : launch.args[at + 1] - if (launchedModel?.includes(':')) { - // OMP 18.17.0 prints `Model "fake/gone:high" not found.` and exits 1 before ACP starts. - const exited = async () => { - throw new AcpAgentExitedError('Model "fake/gone:high" not found.') - } - return { - initialize: exited, - start: exited, - setConfigOption: exited, - deleteSession: exited, - close: async () => {} - } - } - const runtime = new AcpSessionRuntime(agent.stdout, agent.stdin) - return { - initialize: () => runtime.initialize(), - start: (options) => runtime.start(options), - setConfigOption: (id, value) => runtime.setConfigOption(id, value), - deleteSession: (sessionId) => runtime.deleteSession(sessionId), - close: async () => runtime.close() - } - } - })( - { variable: 'PI_CODING_AGENT_DIR', path: '/homes/omp' }, - { workspacePath: '/work/project', ...(selectedModels ? { selectedModels } : {}) } - ) - // A model switch would record usage in the user's OMP account. - expect(agent.frames.map((frame) => frame.method)).toEqual(['initialize', 'session/new']) - return (id: string) => listing.models.find((model) => model.id === id) - } finally { - await rm(dataDirectory, { recursive: true, force: true }) - } -} - -const levels = (model: { efforts: { value: string }[] } | undefined) => - model?.efforts.map((effort) => effort.value) - -it("lists a picked OMP model with its own thinking levels, not the startup model's", async () => { - const listed = await listOmp() - expect(levels(listed('fake/wide'))).toEqual([ - 'off', - 'auto', - 'minimal', - 'low', - 'medium', - 'high', - 'xhigh' - ]) - expect(listed('fake/narrow')).toMatchObject({ efforts: [], effortsKnown: false }) - - const picked = await listOmp(['fake/narrow']) - expect(levels(picked('fake/narrow'))).toEqual(['off', 'auto', 'low', 'high']) - expect(picked('fake/narrow')).not.toHaveProperty('effortsKnown') - expect(picked('fake/wide')).toMatchObject({ efforts: [], effortsKnown: false }) -}) - -it('trusts no menu when OMP cannot start on the picked model', async () => { - const picked = await listOmp(['fake/gone']) - expect(picked('fake/wide')).toMatchObject({ efforts: [], effortsKnown: false }) -}) - -it('still lists OMP when it cannot start on the picked model', async () => { - const picked = await listOmp(['fake/gone:high']) - expect(levels(picked('fake/wide'))).toEqual([ - 'off', - 'auto', - 'minimal', - 'low', - 'medium', - 'high', - 'xhigh' - ]) -}) diff --git a/src/main/acp/fixtures/omp-v18.17.0-model-efforts.json b/src/main/acp/fixtures/omp-v18.17.0-model-efforts.json deleted file mode 100644 index 4ba4f7835a0..00000000000 --- a/src/main/acp/fixtures/omp-v18.17.0-model-efforts.json +++ /dev/null @@ -1,102 +0,0 @@ -{ - "new": { - "configOptions": [ - { - "id": "model", - "name": "Model", - "category": "model", - "type": "select", - "currentValue": "fake/wide", - "options": [ - { "value": "fake/wide", "name": "Wide", "description": "fake/wide" }, - { "value": "fake/narrow", "name": "Narrow", "description": "fake/narrow" }, - { - "value": "apple/on-device", - "name": "Apple AFM 3 Core Advanced", - "description": "apple/on-device" - } - ] - }, - { - "id": "thinking", - "name": "Thinking", - "category": "thought_level", - "type": "select", - "currentValue": "high", - "options": [ - { "value": "off", "name": "Off" }, - { "value": "auto", "name": "Auto", "description": "Auto-detect per prompt" }, - { "value": "minimal", "name": "minimal" }, - { "value": "low", "name": "low" }, - { "value": "medium", "name": "medium" }, - { "value": "high", "name": "high" }, - { "value": "xhigh", "name": "xhigh" } - ] - } - ] - }, - "narrow": { - "configOptions": [ - { - "id": "model", - "name": "Model", - "category": "model", - "type": "select", - "currentValue": "fake/narrow", - "options": [ - { "value": "fake/wide", "name": "Wide", "description": "fake/wide" }, - { "value": "fake/narrow", "name": "Narrow", "description": "fake/narrow" }, - { - "value": "apple/on-device", - "name": "Apple AFM 3 Core Advanced", - "description": "apple/on-device" - } - ] - }, - { - "id": "thinking", - "name": "Thinking", - "category": "thought_level", - "type": "select", - "currentValue": "high", - "options": [ - { "value": "off", "name": "Off" }, - { "value": "auto", "name": "Auto", "description": "Auto-detect per prompt" }, - { "value": "low", "name": "low" }, - { "value": "high", "name": "high" } - ] - } - ] - }, - "unresolved": { - "configOptions": [ - { - "id": "model", - "name": "Model", - "category": "model", - "type": "select", - "currentValue": "fake/wide", - "options": [ - { "value": "fake/wide", "name": "Wide", "description": "fake/wide" }, - { "value": "fake/narrow", "name": "Narrow", "description": "fake/narrow" }, - { - "value": "apple/on-device", - "name": "Apple AFM 3 Core Advanced", - "description": "apple/on-device" - } - ] - }, - { - "id": "thinking", - "name": "Thinking", - "category": "thought_level", - "type": "select", - "currentValue": "off", - "options": [ - { "value": "off", "name": "Off" }, - { "value": "auto", "name": "Auto", "description": "Auto-detect per prompt" } - ] - } - ] - } -} diff --git a/src/main/acp/timeline/acp-timeline-opencode-subagents.test.ts b/src/main/acp/timeline/acp-timeline-opencode-subagents.test.ts index c0c79881234..dce3a8ca956 100644 --- a/src/main/acp/timeline/acp-timeline-opencode-subagents.test.ts +++ b/src/main/acp/timeline/acp-timeline-opencode-subagents.test.ts @@ -2,48 +2,11 @@ import { afterEach, describe, expect, it } from 'vitest' import type { AgentJournalRenderItem } from '../../../shared/agent-session/journal/agent-session-journal-types' import { closeProviderTimelineRigs } from '../../native-chat/agent-session-timeline/provider-timeline-assembler-test-support' import { OPENCODE_ACP_DIALECT } from '../acp-dialects/opencode-dialect' -import { AcpTimelineTranslator } from './acp-timeline-translator' -import { - openAcpFixtureRig, - readAcpFixture, - type AcpFixtureFrame -} from './testing/acp-timeline-fixture.test-support' + +import { openAcpFixtureRig } from './testing/acp-timeline-fixture.test-support' afterEach(closeProviderTimelineRigs) -type Fixture = Awaited> - -function subagents(fixture: Fixture) { - return fixture.subagents().map((view) => ({ - id: view.providerId, - label: view.description, - membership: view.membership, - outcome: view.outcome - })) -} - -function tools(rows: AgentJournalRenderItem[]) { - return rows.flatMap((row) => - row.body.kind === 'tool-call' - ? [{ agentId: row.agentId, name: row.body.name, state: row.body.state }] - : [] - ) -} - -function commands(rows: AgentJournalRenderItem[], agentId: string) { - return rows.flatMap((row) => - row.agentId === agentId && row.body.kind === 'tool-call' - ? [ - typeof row.body.input === 'object' && - row.body.input !== null && - 'command' in row.body.input - ? row.body.input.command - : undefined - ] - : [] - ) -} - function texts(rows: AgentJournalRenderItem[], agentId: string) { return rows.flatMap((row) => row.agentId === agentId && row.body.kind === 'message' @@ -52,146 +15,9 @@ function texts(rows: AgentJournalRenderItem[], agentId: string) { ) } -/** Frames through the `index`-th one that matches. */ -function through(frames: AcpFixtureFrame[], matches: (text: string) => boolean, index = 0) { - const ends = frames.flatMap((frame, at) => (matches(JSON.stringify(frame)) ? [at] : [])) - return frames.slice(0, ends[index] + 1) -} - const opened = () => openAcpFixtureRig(OPENCODE_ACP_DIALECT, { journalOwnership: true }) describe('OpenCode subagents through the shared timeline', () => { - it('a subagent is one record that completes, and its own work is filed under it', async () => { - const fixture = await opened() - const rows = await fixture.feed(await readAcpFixture('opencode-v2-subagent-foreground')) - expect(subagents(fixture)).toEqual([ - { id: 'session-2', label: 'Run sleep command', membership: 'settled', outcome: 'succeeded' } - ]) - expect(tools(rows)).toEqual([ - { agentId: undefined, name: 'subagent', state: 'completed' }, - { agentId: 'session-2', name: 'shell', state: 'completed' } - ]) - expect(texts(rows, 'session-2')).toEqual(['DONE']) - }) - - it('a background subagent stays open after the turn ends and its later work still lands', async () => { - const fixture = await opened() - const frames = await readAcpFixture('opencode-v2-subagent-background') - const turn = through(frames, (text) => text.includes('"stopReason"')) - await fixture.feed(turn) - expect(subagents(fixture)).toMatchObject([{ id: 'session-2', membership: 'live' }]) - // Commit each frame before the next, so later tool updates read their placement back. - for (const frame of frames.slice(turn.length)) { - await fixture.feed([frame]) - } - const rows = await fixture.rig.rows() - expect(subagents(fixture)).toMatchObject([ - { id: 'session-2', membership: 'settled', outcome: 'succeeded' } - ]) - expect(tools(rows).filter((tool) => tool.agentId === 'session-2')).toEqual([ - { agentId: 'session-2', name: 'shell', state: 'completed' } - ]) - expect(commands(rows, 'session-2')).toEqual(['sleep 20']) - expect(texts(rows, 'session-2')).toEqual(['DONE']) - }) - - it('a subagent command that spans the end of the turn keeps running and then completes', async () => { - const fixture = await opened() - const frames = await readAcpFixture('opencode-v2-subagent-background') - const ending = frames.findIndex((frame) => JSON.stringify(frame).includes('"stopReason"')) - const childStart = frames.findIndex((frame) => - JSON.stringify(frame).includes('"sessionUpdate":"tool_call_update","toolCallId":"session-2:') - ) - // The turn ends while the subagent's command runs. - const spanning = [ - ...frames.slice(0, ending), - ...frames.slice(ending + 1, childStart + 1), - frames[ending], - ...frames.slice(childStart + 1) - ] - const rows = await fixture.feed(spanning) - expect(tools(rows).filter((tool) => tool.agentId === 'session-2')).toEqual([ - { agentId: 'session-2', name: 'shell', state: 'completed' } - ]) - }) - - it('Stop interrupts the subagent and settles its running command', async () => { - const fixture = await opened() - const rows = await fixture.feed(await readAcpFixture('opencode-v2-subagent-stop')) - expect(subagents(fixture)).toMatchObject([ - { id: 'session-2', membership: 'settled', outcome: 'cancelled' } - ]) - expect( - tools(rows).filter((tool) => tool.agentId === 'session-2' && tool.state === 'running') - ).toEqual([]) - }) - - it('closing a background subagent after the turn ends interrupts it and leaves nothing running', async () => { - const fixture = await opened() - const frames = await readAcpFixture('opencode-v2-subagent-close') - const closing = frames.findIndex((frame) => frame.message.method === 'session/close') - await fixture.feed(frames.slice(0, closing)) - expect(fixture.lane().children.live()).toEqual(['session-2']) - const rows = await fixture.feed(frames.slice(closing)) - expect(subagents(fixture)).toMatchObject([ - { id: 'session-2', membership: 'settled', outcome: 'cancelled' } - ]) - expect(fixture.lane().children.live()).toEqual([]) - expect( - tools(rows).filter((tool) => tool.agentId === 'session-2' && tool.state === 'running') - ).toEqual([]) - }) - - it('a follow-up to a finished subagent reopens it until its second run ends', async () => { - const fixture = await opened() - const frames = await readAcpFixture('opencode-v2-subagent-rerun') - const running = (text: string) => text.includes('"status":"running"') - const firstRun = through(frames, running, 1) - await fixture.feed(firstRun) - expect(subagents(fixture)).toMatchObject([{ id: 'session-2', membership: 'live' }]) - const rows = await fixture.feed(frames.slice(firstRun.length)) - expect(subagents(fixture)).toMatchObject([ - { id: 'session-2', membership: 'settled', outcome: 'succeeded' } - ]) - expect(texts(rows, 'session-2')).toEqual(['FIRST', 'SECOND']) - }) - - it('a long-running subagent keeps one message open per channel, however many steps it takes', () => { - // OpenCode starts new message ids each step and never appends to an earlier step's. - const mostOpen = (steps: number) => { - const lane = new AcpTimelineTranslator({ - sessionId: 'session-1', - dialect: OPENCODE_ACP_DIALECT - }) - const open = new Set() - let most = 0 - const report = (body: Record) => { - for (const event of lane.notification( - 'opencode/session/child_update', - { ...CHILD, ...body }, - 1000 - )) { - const key = 'item' in event ? JSON.stringify(event.item) : '' - if (event.type === 'text.delta') { - open.add(key) - } else if (event.type === 'text.close') { - open.delete(key) - } - most = Math.max(most, open.size) - } - } - report({ type: 'status', status: 'running' }) - for (let step = 0; step < steps; step++) { - report(chunk('agent_thought_chunk', `msg-${step}:reasoning:0`, 'thinking')) - report(chunk('agent_message_chunk', `msg-${step}`, `step ${step}`)) - } - report({ type: 'status', status: 'completed' }) - return { most, left: open.size } - } - expect(mostOpen(10)).toEqual({ most: 2, left: 0 }) - expect(mostOpen(100)).toEqual({ most: 2, left: 0 }) - }) - it('a reply that resumes after the subagent thinks or runs a tool keeps all of its text', async () => { const fixture = await opened() const lane = fixture.lane() diff --git a/src/main/environment-providers/environment-provider-driver-contract.test-support.ts b/src/main/environment-providers/environment-provider-driver-contract.test-support.ts deleted file mode 100644 index 6dce4869236..00000000000 --- a/src/main/environment-providers/environment-provider-driver-contract.test-support.ts +++ /dev/null @@ -1,74 +0,0 @@ -// Shared lifecycle contract every EnvironmentProviderDriver must pass; drivers call it from their test. -import { describe, expect, it } from 'vitest' -import { - getEphemeralVmRecipeResultConnection, - getEphemeralVmRecipeResultProjectRoot, - isAbsoluteRuntimePath -} from '../../shared/ephemeral-vm-recipes' -import type { - EnvironmentCreateSpec, - EnvironmentProviderDriver, - EnvironmentReady -} from './environment-provider-driver' - -export type EnvironmentProviderDriverContractHarness = { - driver: EnvironmentProviderDriver - /** Ids the backing provider still holds; destroy must leave none behind. */ - liveEnvironmentIds: () => Promise - /** A fresh driver over the same backing provider, as after an Orca restart. */ - reopenDriver: () => EnvironmentProviderDriver -} - -// Handles are persisted with the runtime record, so later calls only ever see the JSON form. -function reload(handle: T): T { - return JSON.parse(JSON.stringify(handle)) -} - -export function runEnvironmentProviderDriverContract( - name: string, - makeHarness: () => Promise, - spec: (workspaceId: string) => EnvironmentCreateSpec = (workspaceId) => ({ workspaceId }) -): void { - describe(`${name} satisfies the environment provider driver contract`, () => { - function expectReady(driver: EnvironmentProviderDriver, ready: EnvironmentReady): void { - expect(ready.handle.driverId).toBe(driver.id) - expect(ready.handle.environmentId).not.toBe('') - expect(['orca-server', 'ssh']).toContain( - getEphemeralVmRecipeResultConnection(ready.result).type - ) - expect(isAbsoluteRuntimePath(getEphemeralVmRecipeResultProjectRoot(ready.result))).toBe(true) - if (driver.capabilities.maxSessionMs !== null) { - expect(ready.session.deadlineAt).not.toBeNull() - expect(ready.session.deadlineAt! - ready.session.startedAt).toBeLessThanOrEqual( - driver.capabilities.maxSessionMs - ) - } - } - - it('creates, suspends, resumes and destroys one environment without leaking it', async () => { - const { driver, liveEnvironmentIds, reopenDriver } = await makeHarness() - const phases: string[] = [] - const created = await driver.create(spec('ws-contract-1'), { - onProgress: ({ phase }) => phases.push(phase) - }) - expectReady(driver, created) - expect(phases.length).toBeGreaterThan(0) - expect(await liveEnvironmentIds()).toContain(created.handle.environmentId) - expect((await driver.status(created.handle)).status).toBe('running') - - const suspended = await driver.suspend(created.handle) - expect(suspended.status).toBe('suspended') - expect((await driver.status(created.handle)).status).toBe('suspended') - - const resumed = await driver.resume(created.handle) - expectReady(driver, resumed) - expect(resumed.handle.environmentId).toBe(created.handle.environmentId) - expect((await driver.status(created.handle)).status).toBe('running') - - await driver.destroy(created.handle) - expect(await liveEnvironmentIds()).not.toContain(created.handle.environmentId) - // Idempotent: cleanup retries after a lost response or a restart must not fail. - await reopenDriver().destroy(reload(created.handle)) - }) - }) -} diff --git a/src/main/environment-providers/shell-recipe-driver.test.ts b/src/main/environment-providers/shell-recipe-driver.test.ts deleted file mode 100644 index d7cafb409b5..00000000000 --- a/src/main/environment-providers/shell-recipe-driver.test.ts +++ /dev/null @@ -1,68 +0,0 @@ -import { mkdirSync, mkdtempSync, readdirSync, rmSync, writeFileSync } from 'node:fs' -import { tmpdir } from 'node:os' -import { join } from 'node:path' -import { afterEach } from 'vitest' -import { encodePairingOffer } from '../../shared/remote-runtime/pairing/pairing' -import { PAIRING_OFFER_VERSION } from '../../shared/remote-runtime/pairing/mobile-relay-pairing-offer' -import type { OrcaVmRecipe } from '../../shared/orca-yaml-hook-types' -import { runEnvironmentProviderDriverContract } from './environment-provider-driver-contract.test-support' -import { ShellRecipeDriver } from './shell-recipe-driver' - -const tmpRoots: string[] = [] - -afterEach(() => { - for (const root of tmpRoots.splice(0)) { - rmSync(root, { recursive: true, force: true }) - } -}) - -// A recipe whose "VMs" are files in a state dir, so the contract can see what destroy left behind. -function makeFileBackedRecipe(): { repoPath: string; stateDir: string; recipe: OrcaVmRecipe } { - const repoPath = mkdtempSync(join(tmpdir(), 'orca-shell-recipe-driver-')) - tmpRoots.push(repoPath) - const stateDir = join(repoPath, 'vms') - mkdirSync(stateDir) - const pairingCode = encodePairingOffer({ - v: PAIRING_OFFER_VERSION, - endpoint: 'wss://vm.example.com', - deviceToken: 'token', - publicKeyB64: 'public-key' - }) - const vmFile = `require('node:path').join(${JSON.stringify(stateDir)}, process.env.ORCA_VM_INSTANCE_ID)` - const printResult = `console.log(JSON.stringify({ schemaVersion: 1, pairingCode: ${JSON.stringify(pairingCode)}, projectRoot: '/workspace/repo' }))` - const scripts = { - create: `require('node:fs').writeFileSync(${vmFile}, 'running'); ${printResult}`, - suspend: `require('node:fs').writeFileSync(${vmFile}, 'suspended')`, - // Like a real provider: a destroyed VM cannot be woken. - resume: `const fs = require('node:fs'); if (!fs.existsSync(${vmFile})) process.exit(1); fs.writeFileSync(${vmFile}, 'running'); ${printResult}`, - destroy: `require('node:fs').rmSync(${vmFile}, { force: true })` - } - const commands = Object.fromEntries( - Object.entries(scripts).map(([mode, source]) => { - const scriptPath = join(repoPath, `${mode}.js`) - writeFileSync(scriptPath, source) - return [mode, `"${process.execPath}" "${scriptPath}"`] - }) - ) - return { - repoPath, - stateDir, - recipe: { - id: 'file-backed', - name: 'File-backed VM', - create: commands.create, - suspend: commands.suspend, - resume: commands.resume, - destroy: commands.destroy - } - } -} - -runEnvironmentProviderDriverContract('ShellRecipeDriver', async () => { - const { repoPath, stateDir, recipe } = makeFileBackedRecipe() - return { - driver: new ShellRecipeDriver(recipe, repoPath), - liveEnvironmentIds: async () => readdirSync(stateDir), - reopenDriver: () => new ShellRecipeDriver(recipe, repoPath) - } -}) diff --git a/src/main/runtime/terminal/stop/launch-evidence-resume-records.test.ts b/src/main/runtime/terminal/stop/launch-evidence-resume-records.test.ts deleted file mode 100644 index a6783edbad3..00000000000 --- a/src/main/runtime/terminal/stop/launch-evidence-resume-records.test.ts +++ /dev/null @@ -1,115 +0,0 @@ -// A pane's saved recipe follows main's word on its launch, not the session the agent names. -import { describe, expect, it, vi } from 'vitest' -import type { AgentStatusIpcPayload } from '../../../../shared/agent-status/contracts/agent-status-ipc-payload' -import type { SleepingAgentSessionRecord } from '../../../../shared/agent-session/lifecycle/agent-session-resume' -import type { ExecutionHostId } from '../../../../shared/execution-host' -import type { LayoutApplyOutcome } from '../../../../shared/workspace-layout/workspace-layout-command-types' -import { applyCommandToSession } from '../../../persistence/terminal-topology/terminal-pane-admission' - -vi.mock('../../../../shared/layout-owner-switch', () => ({ ORCA_LAYOUT_OWNER_IS_MAIN: true })) - -// Preserve side-effect order: mocks, lifecycle, fixtures. -await import('../../orca-runtime-tests/runtime-test-module-registration.spec') -const { OrcaRuntimeService } = await import('../../orca-runtime') -await import('../../orca-runtime-tests/orca-runtime-test-lifecycle.spec') -const { - HEADLESS_LEAF_ID, - TEST_WORKTREE_ID, - makeRuntimeStoreWithWorkspaceSession, - makeWorkspaceSessionWithHeadlessTerminal -} = await import('../../orca-runtime-tests/orca-runtime-test-fixtures.spec') - -const PANE_KEY = `host-tab:${HEADLESS_LEAF_ID}` -const PTY_ID = 'persisted-pty' -const SESSION = { key: 'session_id' as const, id: 'claude-session-1' } -const RECIPE = { agentArgs: '--dangerously-skip-permissions', agentEnv: {} } - -function savedRecord(origin: SleepingAgentSessionRecord['origin']): SleepingAgentSessionRecord { - return { - paneKey: PANE_KEY, - tabId: 'host-tab', - worktreeId: TEST_WORKTREE_ID, - agent: 'claude', - providerSession: SESSION, - prompt: '', - state: 'done', - capturedAt: 5, - updatedAt: 5, - origin, - connectionId: null, - launchConfig: RECIPE - } -} - -function claudeRow(): AgentStatusIpcPayload { - return { - paneKey: PANE_KEY, - worktreeId: TEST_WORKTREE_ID, - agentType: 'claude', - providerSession: SESSION, - state: 'working', - prompt: 'keep going', - receivedAt: 10, - stateStartedAt: 10, - connectionId: null - } -} - -function makeRuntime(saved: SleepingAgentSessionRecord) { - const { runtimeStore, getSession, setSession } = makeRuntimeStoreWithWorkspaceSession( - makeWorkspaceSessionWithHeadlessTerminal({ - sleepingAgentSessionsByPaneKey: { [PANE_KEY]: saved } - }) - ) - const applyLayoutCommand = vi.fn( - async ( - command: Parameters[2], - hostIds: ExecutionHostId[] - ): Promise => { - for (const hostId of hostIds) { - const change = applyCommandToSession(hostId, getSession(), command) - if (change.session) { - setSession(change.session) - } - } - return { ok: true, result: {} } - } - ) - const runtime = new OrcaRuntimeService(Object.assign(runtimeStore, { applyLayoutCommand })) - runtime.syncWindowGraph(0, { tabs: [], leaves: [] }) - const binding = { tabId: 'host-tab', leafId: HEADLESS_LEAF_ID, incarnationId: 'p1' } - return { - runtime, - binding, - getSession, - savedAfterStatus: async () => { - runtime.recordHookStatusResume(claudeRow()) - await runtime['hookStatusResumeWrites'] - return getSession().sleepingAgentSessionsByPaneKey?.[PANE_KEY] - } - } -} - -describe('a pane saved recipe and the launch running in it', () => { - it('gives an agent typed into a fresh shell none of the session earlier recipe', async () => { - const { runtime, binding, savedAfterStatus } = makeRuntime(savedRecord('live')) - runtime.registerPty(PTY_ID, TEST_WORKTREE_ID, null, { ...binding, freshSpawn: true }) - const saved = await savedAfterStatus() - expect(saved?.state).toBe('working') - expect(saved?.launchConfig).toBeUndefined() - }) - - it('keeps a sleep snapshot for its wake without lending its recipe to a typed resume', async () => { - const { runtime, binding, getSession, savedAfterStatus } = makeRuntime( - savedRecord('worktree-sleep') - ) - runtime.registerPty(PTY_ID, TEST_WORKTREE_ID, null, { - ...binding, - agentLaunchAuthority: { launchToken: 'launch-1', launchAgent: 'claude', launchConfig: RECIPE } - }) - runtime.emitDaemonPtyTransientFact(PTY_ID, { kind: 'command-finished', exitCode: 0 }) - expect(await savedAfterStatus()).toEqual(savedRecord('worktree-sleep')) - Object.assign(runtime, { getAgentProviderSessionSnapshotFn: () => [claudeRow()] }) - expect(runtime.quitResumeRecordsFor(getSession())).toEqual([]) - }) -}) diff --git a/src/main/speech/stt-vad-pause-segmenter.test.ts b/src/main/speech/stt-vad-pause-segmenter.test.ts index 4ad6cd1efdb..a5e8285fb1e 100644 --- a/src/main/speech/stt-vad-pause-segmenter.test.ts +++ b/src/main/speech/stt-vad-pause-segmenter.test.ts @@ -59,36 +59,6 @@ describe('VadPauseSegmenter', () => { expect(framesOf(await segmenter.flush())).toEqual([[0.11, 0.12, 0.13, 0.14, 0.7, 0.6]]) }) - it('keeps a phrase whole across a pause shorter than 0.6 s', async () => { - const segmenter = new VadPauseSegmenter(fakeVad(), RATE, FRAME) - const ready = await segmenter.push( - audio([0.9, 0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2]) - ) - expect(framesOf(ready)).toEqual([ - [0.9, 0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2] - ]) - }) - - it('holds a click shorter than 0.25 s of speech until real speech joins it, without opening a phrase', async () => { - const segmenter = new VadPauseSegmenter(fakeVad(), RATE, FRAME) - expect(await segmenter.push(audio([0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2]))).toEqual([]) - expect(segmenter.phraseOpen).toBe(false) - expect(await segmenter.push(audio([0.8, 0.8]))).toEqual([]) - expect(segmenter.phraseOpen).toBe(true) - const ready = await segmenter.push(audio([0.2, 0.2, 0.2, 0.2, 0.2, 0.2])) - expect(framesOf(ready)).toEqual([ - [0.9, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.8, 0.8, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2] - ]) - expect(segmenter.phraseOpen).toBe(false) - }) - - it('returns nothing for a dictation that never had speech', async () => { - const segmenter = new VadPauseSegmenter(fakeVad(), RATE, FRAME) - expect(await segmenter.push(audio([0.1, 0.2, 0.3, 0.1]))).toEqual([]) - expect(await segmenter.flush()).toEqual([]) - expect(segmenter.heldSamples).toBe(0) - }) - it('keeps every held sample when the detector fails mid-dictation', async () => { const vad = fakeVad() const segmenter = new VadPauseSegmenter( diff --git a/src/renderer/src/components/automations/runs/use-automation-runs-dashboard-payload.test.tsx b/src/renderer/src/components/automations/runs/use-automation-runs-dashboard-payload.test.tsx deleted file mode 100644 index bd2e16684ec..00000000000 --- a/src/renderer/src/components/automations/runs/use-automation-runs-dashboard-payload.test.tsx +++ /dev/null @@ -1,151 +0,0 @@ -// @vitest-environment happy-dom - -import { act } from 'react' -import { createRoot, type Root } from 'react-dom/client' -import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' -import type { AutomationRun } from '../../../../../shared/automations/history/automation-run' -import { makeAutomationListRow, makeRun } from '../automations-page-fixtures' -import * as dispatch from '../hosts/automation-row-action-dispatch' -import { useAutomationRunsDashboard } from './use-automation-runs-dashboard' - -vi.mock('../hosts/automation-row-action-dispatch', async (importOriginal) => ({ - ...(await importOriginal()), - dispatchAutomationRunHistoryPage: vi.fn(), - dispatchAutomationRunHistory: vi.fn() -})) - -const pageSpy = vi.mocked(dispatch.dispatchAutomationRunHistoryPage) -const selectedSpy = vi.mocked(dispatch.dispatchAutomationRunHistory) -const row = makeAutomationListRow() -const rows = [row] -const context = { capturedOwners: new Map(), authority: { kind: 'desktop' as const } } -const legacyTarget = () => null -const authorityForRow = () => ({ kind: 'desktop' as const }) -type DashboardResult = ReturnType -type PageResult = Awaited> - -let root: Root -let container: HTMLDivElement -let latest: DashboardResult | null = null - -const frames: { enabled: boolean; ids: string[]; loading: boolean }[] = [] - -function Harness({ enabled }: { enabled: boolean; detail?: boolean }): null { - latest = useAutomationRunsDashboard({ - enabled, - rows, - context, - legacyTarget, - authorityForRow, - reloadToken: 0 - }) - - frames.push({ - enabled, - ids: latest.entries.map((entry) => entry.run.id), - loading: latest.loading - }) - return null -} - -async function render(enabled: boolean, detail: boolean = false): Promise { - await act(async () => { - root.render() - }) -} - -function entries(): DashboardResult['entries'] { - if (!latest) { - throw new Error('Expected the mounted dashboard result') - } - return latest.entries -} - -function fullRun(id: string, scheduledFor: number = 10): AutomationRun { - return makeRun({ - id, - scheduledFor, - title: `Run ${id}`, - trigger: 'manual', - outputSnapshot: { - format: 'plain_text', - content: `Terminal output for ${id}\n`, - capturedAt: 20, - truncated: false - } - }) -} - -function pendingPage() { - let complete: ((result: PageResult) => void) | null = null - const promise = new Promise((resolve) => { - complete = resolve - }) - return { - promise, - resolve: (result: PageResult): void => { - if (!complete) { - throw new Error('Expected a pending history request') - } - complete(result) - } - } -} - -beforeEach(() => { - globalThis.IS_REACT_ACT_ENVIRONMENT = true - container = document.createElement('div') - document.body.append(container) - root = createRoot(container) - frames.length = 0 -}) - -afterEach(() => { - act(() => root.unmount()) - container.remove() - latest = null - - frames.length = 0 - pageSpy.mockReset() - selectedSpy.mockReset() -}) - -describe('automation dashboard run payload ownership', () => { - it('preserves the first duplicate, chronological order and cursor retirement when paging', async () => { - pageSpy - .mockResolvedValueOnce({ - ok: true, - value: { runs: [fullRun('head', 20)], nextCursor: 'next-page' } - }) - .mockResolvedValueOnce({ - ok: true, - value: { runs: [fullRun('head', 30), fullRun('older', 10)], nextCursor: null } - }) - await render(true) - await act(async () => latest?.loadMore()) - - expect(entries().map((entry) => [entry.run.id, entry.run.scheduledFor])).toEqual([ - ['head', 20], - ['older', 10] - ]) - expect(pageSpy.mock.calls[1]?.[2].cursor).toBe('next-page') - expect(latest?.hasMore).toBe(false) - }) - - it('ignores a closed request that finishes after a fresh re-entry', async () => { - const pending = pendingPage() - pageSpy.mockReturnValueOnce(pending.promise).mockResolvedValueOnce({ - ok: true, - value: { runs: [fullRun('fresh')], nextCursor: null } - }) - await render(true) - await render(false) - await render(true) - await act(async () => { - pending.resolve({ ok: true, value: { runs: [fullRun('obsolete')], nextCursor: 'stale' } }) - }) - - expect(entries().map((entry) => entry.run.id)).toEqual(['fresh']) - expect(latest?.hasMore).toBe(false) - }) -}) diff --git a/src/renderer/src/components/automations/runs/use-automation-runs-dashboard.test.tsx b/src/renderer/src/components/automations/runs/use-automation-runs-dashboard.test.tsx deleted file mode 100644 index 8ebe3f8a7e3..00000000000 --- a/src/renderer/src/components/automations/runs/use-automation-runs-dashboard.test.tsx +++ /dev/null @@ -1,131 +0,0 @@ -// @vitest-environment happy-dom - -import { act, useCallback } from 'react' -import { createRoot, type Root } from 'react-dom/client' -import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' -import type { AutomationAuthorityRef } from '../../../../../shared/automations/ownership/automation-owner-ref' -import type { AutomationRunsPage } from '../../../../../shared/automations/history/automation-run' -import type { AutomationHostTarget } from '../hosts/automation-host-client' -import { makeAutomationListRow, makeRun } from '../automations-page-fixtures' -import * as dispatch from '../hosts/automation-row-action-dispatch' -import { useAutomationRunsDashboard } from './use-automation-runs-dashboard' - -vi.mock('../hosts/automation-row-action-dispatch', async (importOriginal) => ({ - ...(await importOriginal()), - dispatchAutomationRunHistoryPage: vi.fn() -})) - -const historySpy = vi.mocked(dispatch.dispatchAutomationRunHistoryPage) - -const PAGES: Record = { - head: { runs: [makeRun({ id: 'run-2', createdAt: 2 })], nextCursor: '2:run-2' }, - '2:run-2': { runs: [makeRun({ id: 'run-1', createdAt: 1 })], nextCursor: null } -} - -const row = makeAutomationListRow() -const rows = [row] -const context = { capturedOwners: new Map(), authority: { kind: 'desktop' as const } } - -type DashboardResult = ReturnType - -let container: HTMLDivElement -let root: Root -let latest: DashboardResult | null = null - -type HarnessProps = { - enabled: boolean - authority?: AutomationAuthorityRef - target?: AutomationHostTarget | null - currentRows?: typeof rows -} - -function Harness({ enabled, authority, target, currentRows = rows }: HarnessProps): null { - latest = useAutomationRunsDashboard({ - enabled, - rows: currentRows, - context, - legacyTarget: useCallback(() => target ?? null, [target]), - authorityForRow: useCallback(() => authority ?? { kind: 'desktop' }, [authority]), - reloadToken: 0 - }) - return null -} - -async function render(enabled: boolean, props: Omit = {}): Promise { - await act(async () => { - root.render() - }) -} - -beforeEach(() => { - historySpy.mockImplementation(async (_context, _row, options) => ({ - ok: true, - value: PAGES[options.cursor ?? 'head'] - })) - container = document.createElement('div') - document.body.append(container) - root = createRoot(container) -}) - -afterEach(() => { - act(() => root.unmount()) - container.remove() - latest = null - historySpy.mockReset() -}) - -describe('useAutomationRunsDashboard', () => { - it('keeps the cursor when a load more fails so the page stays retryable', async () => { - await render(true) - historySpy.mockImplementationOnce(async () => ({ - ok: false, - notice: { message: 'offline', recovery: null, severity: 'failure' } - })) - - await act(async () => latest?.loadMore()) - expect(latest?.entries.map((entry) => entry.run.id)).toEqual(['run-2']) - expect(latest?.hasMore).toBe(true) - - await act(async () => latest?.loadMore()) - - expect(historySpy.mock.calls.at(-1)?.[2].cursor).toBe('2:run-2') - expect(latest?.entries.map((entry) => entry.run.id)).toEqual(['run-2', 'run-1']) - }) - - it('reloads from the head when the authority is re-paired', async () => { - const paired = (pairingRevision: number): AutomationAuthorityRef => ({ - kind: 'runtime', - environmentId: 'env-1', - pairingRevision - }) - await render(true, { authority: paired(1) }) - await act(async () => latest?.loadMore()) - expect(latest?.entries).toHaveLength(2) - historySpy.mockClear() - - await render(true, { authority: paired(2) }) - - expect(historySpy.mock.calls.at(-1)?.[2].cursor).toBeUndefined() - expect(latest?.entries.map((entry) => entry.run.id)).toEqual(['run-2']) - }) - - it('reloads from the head when an uncaptured row moves to another fallback target', async () => { - await render(true, { target: { kind: 'local' } }) - await act(async () => latest?.loadMore()) - expect(latest?.entries).toHaveLength(2) - historySpy.mockClear() - - await render(true, { target: { kind: 'environment', environmentId: 'env-1' } }) - - expect(historySpy.mock.calls.at(-1)?.[2].cursor).toBeUndefined() - expect(latest?.entries.map((entry) => entry.run.id)).toEqual(['run-2']) - }) - it('updates the host label on existing run history when the host row changes', async () => { - await render(true, { currentRows: [{ ...row, hostLabel: 'previous name' }] }) - expect(latest?.entries[0]?.run.id).toBe('run-2') - await render(true, { currentRows: [{ ...row, hostLabel: 'new name' }] }) - expect(latest?.entries[0]?.run.id).toBe('run-2') - expect(latest?.entries[0]?.row.hostLabel).toBe('new name') - expect(latest?.entries[0]?.searchText).toContain('new name') - }) -}) diff --git a/src/renderer/src/components/native-chat/session/status/structured-agent-session-status-bridge-tab-actions.test.tsx b/src/renderer/src/components/native-chat/session/status/structured-agent-session-status-bridge-tab-actions.test.tsx index 61bf01b6e34..52f9f5a6060 100644 --- a/src/renderer/src/components/native-chat/session/status/structured-agent-session-status-bridge-tab-actions.test.tsx +++ b/src/renderer/src/components/native-chat/session/status/structured-agent-session-status-bridge-tab-actions.test.tsx @@ -7,10 +7,7 @@ import type { AgentSessionStatusEvent, AgentSessionStatusSummary } from '../../../../../../shared/agent-session/agent-session-wire' -import { - LOCAL_EXECUTION_HOST_ID, - type ExecutionHostId -} from '../../../../../../shared/execution-host' + import type { Tab } from '../../../../../../shared/tab-types' import type { AppState } from '@/store/types' @@ -27,13 +24,11 @@ const mocks = vi.hoisted(() => { subscribeStatus: ReturnType supportsCapability: ReturnType unsubscribe: ReturnType - applyStatus: ReturnType void>> } = { store: null, subscribeStatus: vi.fn(), supportsCapability: vi.fn(), - unsubscribe: vi.fn(), - applyStatus: vi.fn() + unsubscribe: vi.fn() } return hoisted }) @@ -41,19 +36,6 @@ const mocks = vi.hoisted(() => { vi.mock('@/store', async () => { const { createTestStore } = await import('@/store/testing/app-store-test-fixture') const useAppStore = createTestStore() - const { transactAgentStatuses } = useAppStore.getState() - useAppStore.setState({ - transactAgentStatuses: (operation) => - transactAgentStatuses((transaction) => - operation({ - ...transaction, - apply: (update) => { - mocks.applyStatus(update) - return transaction.apply(update) - } - }) - ) - }) mocks.store = useAppStore return { useAppStore } }) @@ -103,16 +85,6 @@ function statusFor(tabId: string, sessionId: string) { ] } -function openChat(id: string, sessionId: string, executionHostId?: ExecutionHostId): void { - useAppStore.getState().createUnifiedTab('wt-1', 'agent-session', { - id, - entityId: sessionId, - agentSessionAgent: 'codex', - label: 'Codex Chat', - ...(executionHostId ? { executionHostId } : {}) - }) -} - // Real tab actions annotate their writes, so these reach the bridge's addressed reconciliation. describe('StructuredAgentSessionStatusBridge with tab actions', () => { beforeEach(() => { @@ -136,54 +108,6 @@ describe('StructuredAgentSessionStatusBridge with tab actions', () => { resetStructuredAgentSessionStatusFeedsForTests() }) - it('removes a closed chat status and keeps an unrelated tab write quiet', async () => { - render() - act(() => openChat('chat-1', 'session-1', LOCAL_EXECUTION_HOST_ID)) - await waitFor(() => expect(mocks.subscribeStatus).toHaveBeenCalledOnce()) - act(() => - feed().emit({ type: 'snapshot', sessions: [summary({ status: 'working', updatedAt: 5 })] }) - ) - expect(statusFor('chat-1', 'session-1')).toEqual( - expect.objectContaining({ state: 'working', updatedAt: 5 }) - ) - - act(() => useAppStore.getState().setTabLabel('chat-1', 'Renamed')) - expect(statusFor('chat-1', 'session-1')).toEqual( - expect.objectContaining({ state: 'working', terminalTitle: 'Renamed' }) - ) - - act(() => useAppStore.getState().closeUnifiedTab('chat-1')) - expect(statusFor('chat-1', 'session-1')).toBeUndefined() - }) - - it('retires the old pane when a tab is rebound to another session and ignores its late tick', async () => { - render() - act(() => openChat('chat-1', 'session-1', LOCAL_EXECUTION_HOST_ID)) - await waitFor(() => expect(mocks.subscribeStatus).toHaveBeenCalledOnce()) - act(() => feed().emit({ type: 'snapshot', sessions: [summary()] })) - expect(statusFor('chat-1', 'session-1')).toBeDefined() - - act(() => { - useAppStore.getState().closeUnifiedTab('chat-1') - openChat('chat-1', 'session-2', LOCAL_EXECUTION_HOST_ID) - }) - // The last chat's close released the host feed, so the rebound chat subscribes afresh. - await waitFor(() => expect(mocks.subscribeStatus).toHaveBeenCalledTimes(2)) - act(() => - feed(-1).emit({ - type: 'snapshot', - sessions: [summary({ sessionId: 'session-2', status: 'working', updatedAt: 9 })] - }) - ) - act(() => - feed(-1).emit({ type: 'status', session: summary({ status: 'working', updatedAt: 10 }) }) - ) - expect(statusFor('chat-1', 'session-1')).toBeUndefined() - expect(statusFor('chat-1', 'session-2')).toEqual( - expect.objectContaining({ state: 'working', updatedAt: 9 }) - ) - }) - it('rebinds an unstamped chat to its new owner host without changing membership', async () => { useAppStore.setState({ unifiedTabsByWorktree: { diff --git a/src/renderer/src/hooks/clipboard/primary-selection/usePrimarySelectionPaste.test.tsx b/src/renderer/src/hooks/clipboard/primary-selection/usePrimarySelectionPaste.test.tsx index 28e37fcf1ba..ffcb07b58c7 100644 --- a/src/renderer/src/hooks/clipboard/primary-selection/usePrimarySelectionPaste.test.tsx +++ b/src/renderer/src/hooks/clipboard/primary-selection/usePrimarySelectionPaste.test.tsx @@ -60,26 +60,6 @@ function dispatchMiddleMouseUp(target: HTMLElement): MouseEvent { return event } -function dispatchMiddleAuxClick(target: HTMLElement): MouseEvent { - const event = new MouseEvent('auxclick', { bubbles: true, button: 1, cancelable: true }) - target.dispatchEvent(event) - return event -} - -function dispatchNativePasteBeforeInput(target: HTMLElement): Event { - const event = - typeof InputEvent === 'function' - ? new InputEvent('beforeinput', { - bubbles: true, - cancelable: true, - data: 'native', - inputType: 'insertFromPaste' - }) - : new Event('beforeinput', { bubbles: true, cancelable: true }) - target.dispatchEvent(event) - return event -} - function dispatchMiddleClick(target: HTMLElement): void { dispatchMiddleMouseDown(target) dispatchMiddleMouseUp(target) @@ -127,7 +107,6 @@ afterEach(async () => { document.body.replaceChildren() setUserAgent(originalUserAgent) vi.clearAllMocks() - vi.useRealTimers() }) describe('usePrimarySelectionPaste', () => { @@ -154,69 +133,4 @@ describe('usePrimarySelectionPaste', () => { expect(currentTarget.value).toBe('current') expect(document.activeElement).toBe(currentTarget) }) - - it('owns Linux middle-click paste for editable DOM targets', async () => { - setUserAgent('Mozilla/5.0 (X11; Linux x86_64)') - readPrimarySelectionTextMock.mockResolvedValue('beta') - await renderProbe() - const textarea = appendTextarea('alpha omega') - textarea.focus() - textarea.setSelectionRange(6, 11) - let nativeBeforeInput!: Event - let mouseUp!: MouseEvent - let auxClick!: MouseEvent - - await act(async () => { - dispatchMiddleMouseDown(textarea) - nativeBeforeInput = dispatchNativePasteBeforeInput(textarea) - mouseUp = dispatchMiddleMouseUp(textarea) - auxClick = dispatchMiddleAuxClick(textarea) - await flushPromises() - }) - - expect(nativeBeforeInput.defaultPrevented).toBe(true) - expect(mouseUp.defaultPrevented).toBe(true) - expect(auxClick.defaultPrevented).toBe(true) - expect(readPrimarySelectionTextMock).toHaveBeenCalledTimes(1) - expect(textarea.value).toBe('alpha beta') - }) - - it('does not suppress an unrelated document paste while the terminal window is armed', async () => { - setUserAgent('Mozilla/5.0 (X11; Linux x86_64)') - // Armed terminal suppression must leave unrelated form pastes intact. - consumeNativePasteMock.mockReturnValue(true) - await renderProbe() - const unrelatedTextarea = appendTextarea() - let nativeBeforeInput!: Event - const nativePaste = new Event('paste', { bubbles: true, cancelable: true }) - - await act(async () => { - nativeBeforeInput = dispatchNativePasteBeforeInput(unrelatedTextarea) - unrelatedTextarea.dispatchEvent(nativePaste) - await flushPromises() - }) - - expect(nativeBeforeInput.defaultPrevented).toBe(false) - expect(nativePaste.defaultPrevented).toBe(false) - // Check target before consuming; unrelated pastes must not burn the arm and permit a terminal double-paste. - expect(consumeNativePasteMock).not.toHaveBeenCalled() - }) - - it('does not keep middle-click ownership after the gesture window expires', async () => { - vi.useFakeTimers() - vi.setSystemTime(1_000) - readPrimarySelectionTextMock.mockResolvedValue('late') - await renderProbe() - const textarea = appendTextarea('unchanged') - textarea.focus() - - dispatchMiddleMouseDown(textarea) - vi.setSystemTime(1_751) - const mouseUp = dispatchMiddleMouseUp(textarea) - await flushPromises() - - expect(mouseUp.defaultPrevented).toBe(false) - expect(readPrimarySelectionTextMock).not.toHaveBeenCalled() - expect(textarea.value).toBe('unchanged') - }) }) diff --git a/src/shared/dictation-dictionary.test.ts b/src/shared/dictation-dictionary.test.ts index f7b159cd011..cb72fb380d3 100644 --- a/src/shared/dictation-dictionary.test.ts +++ b/src/shared/dictation-dictionary.test.ts @@ -1,10 +1,5 @@ import { describe, expect, it } from 'vitest' -import { - createDictationDictionary, - dictationTermsFor, - dictationVocabulary -} from './dictation-dictionary' -import { createDictationTextPipeline } from './dictation-text-pipeline' +import { createDictationDictionary } from './dictation-dictionary' const spell = (terms: Parameters[0], text: string): string => createDictationDictionary(terms)!(text) @@ -30,57 +25,4 @@ describe('createDictationDictionary', () => { 'Wrap it in useMemo and pass the onChange call back.' ) }) - - it('replaces what the model hears instead, and fixes the spelling of the term itself', () => { - const terms = [{ term: 'kubectl', heardAs: 'Quebecal' }, { term: 'Orca' }, { term: 'pnpm' }] - expect(spell(terms, 'Use Quebecal to scale it.')).toBe('Use kubectl to scale it.') - expect(spell(terms, "Open or ca's settings and run PNPM.")).toBe( - "Open Orca's settings and run pnpm." - ) - }) - - it('keeps a term with symbols whole when the model already wrote it', () => { - const terms = [{ term: 'C#' }, { term: '.NET' }] - expect(spell(terms, 'Port the C# app to .NET, not plan C.')).toBe( - 'Port the C# app to .NET, not plan C.' - ) - }) - - it('leaves longer words and words split by a sentence end alone', () => { - const terms = [{ term: 'React' }, { term: 'onChange' }] - expect(spell(terms, 'Reactive state. Turn it on. Change it.')).toBe( - 'Reactive state. Turn it on. Change it.' - ) - }) - - it('replaces a "heard as" inside text written without spaces', () => { - expect(spell([{ term: '光輝', heardAs: 'こうき' }], 'こうきさんに送って')).toBe( - '光輝さんに送って' - ) - }) - - it('lets the repo name join its spoken words but never recase a single word', () => { - const oneWord = dictationTermsFor({}, { displayName: 'orca' }) - const words = dictationTermsFor({}, { displayName: 'dd-scratch-repo' }) - - expect(createDictationDictionary(oneWord)).toBeNull() - expect(dictationVocabulary(oneWord)).toEqual(['orca']) - expect(spell(words, 'Open dd scratch repo now.')).toBe('Open dd-scratch-repo now.') - expect(dictationTermsFor({ dictationRepoNameHint: false }, { displayName: 'orca' })).toEqual([]) - }) - - it("prefers a repo's spelling over a personal one", () => { - const terms = dictationTermsFor( - { dictationTerms: [{ term: 'rpc' }] }, - { dictationTerms: [{ term: 'RPC' }] } - ) - expect(spell(terms, 'Call the rpc.')).toBe('Call the RPC.') - }) -}) - -describe('createDictationTextPipeline with terms', () => { - it('spells a lowercase term after a sentence-start filler goes', () => { - const clean = createDictationTextPipeline({ language: 'en' }, {}, [{ term: 'pnpm' }])! - expect(clean('Um, pnpm test.')).toBe('pnpm test.') - }) }) diff --git a/tests/e2e/cross-version-wire/cross-version-stop-note-scenarios.ts b/tests/e2e/cross-version-wire/cross-version-stop-note-scenarios.ts index bb1f8559360..4a483ba849c 100644 --- a/tests/e2e/cross-version-wire/cross-version-stop-note-scenarios.ts +++ b/tests/e2e/cross-version-wire/cross-version-stop-note-scenarios.ts @@ -1,13 +1,10 @@ import { describe, expect, it } from 'vitest' -import { agentJournalItemKey } from '../../../src/shared/agent-session/journal/agent-session-journal-item-key' + import { AGENT_JOURNAL_THREAD_SCOPE } from '../../../src/shared/agent-session/journal/agent-session-journal-types' -import { HISTORY_PAGE_CONTENT_BUDGET_BYTES } from '../../../src/main/native-chat/agent-session-wire/publication/history/agent-session-history-page-bounds' import { createReleasedStopNoteRig, noteId, - noteIdentity, turnItemId, - scope, ordinary, unconfirmed, type ScenarioPort @@ -19,24 +16,7 @@ export function describeReleasedStopNoteProjection(port: ScenarioPort): void { (ref) => { const rig = createReleasedStopNoteRig(port, ref) const { page, subscribe, events, reduce, render, turn, note, seed } = rig - it('renders the derived page with the interrupted turn outside the page', async () => { - await turn('interrupted') - await note() - const snapshot = rig.journal.snapshot() - const stored = rig.journal.itemBody(noteId) - const notePage = await page({ direction: 'tail', limit: 1 }) - expect(notePage.items.map((item) => item.itemId)).toEqual([noteId]) - const state = rig.client.reduceStructuredAgentSession( - rig.client.EMPTY_STRUCTURED_AGENT_SESSION, - { - type: 'history-page', - page: notePage - } - ) - expect(render(state)).toMatchObject({ type: 'text', text: ordinary.text }) - expect(state.items[0]).toEqual(snapshot.items.find((item) => item.itemId === noteId)) - expect(rig.journal.itemBody(noteId)).toBe(stored) - }) + it.each([false, true])( 'renders the same-revision note with a live turn end (lifecycle batch=%s)', async (batch) => { @@ -107,76 +87,6 @@ export function describeReleasedStopNoteProjection(port: ScenarioPort): void { expect(render(state)).toEqual({ type: 'text', text: ordinary.text }) expect(rig.journal.cursor()).toEqual(held) }) - it('replays to a released desktop and pages a fresh snapshot back to the dependent note', async () => { - const before = await seed() - const held = rig.journal.cursor() - for (const clientMessageId of ['large-a', 'large-b']) { - await rig.journal.appendItem( - { provider: 'orca', clientMessageId }, - { - kind: 'message', - role: 'assistant', - blocks: [ - { - type: 'text', - text: 'x'.repeat(Math.floor(HISTORY_PAGE_CONTENT_BUDGET_BYTES * 0.6)) - } - ] - }, - { fence: 1, turnScope: AGENT_JOURNAL_THREAD_SCOPE } - ) - } - await turn('interrupted') - const first = await page({ direction: 'after', cursor: held, limit: 3 }) - expect(first.hasNewer).toBe(true) - const caughtUp = await subscribe(held) - expect(events(caughtUp).every((event) => event.type === 'batch')).toBe(true) - expect(reduce(before, caughtUp).cursor).toEqual(rig.journal.cursor()) - const fresh = await subscribe(null) - let after = reduce(before, fresh) - expect(events(fresh)[0]).toMatchObject({ type: 'snapshot', page: { hasOlder: true } }) - while (after.hasOlder && !after.items.some((item) => item.itemId === noteId)) { - const requestedCursor = { epoch: after.epoch!, sequence: after.items[0]!.sequence } - const older = await page({ direction: 'before', cursor: requestedCursor }) - after = rig.client.reduceStructuredAgentSession(after, { - type: 'older-page', - requestedCursor, - page: older - }) - } - expect(render(after)).toEqual({ type: 'text', text: ordinary.text }) - expect(after.cursor).toEqual(rig.journal.cursor()) - }) - it('preserves removal when the operation note is re-keyed before the interrupted end', async () => { - const op = { provider: 'orca', clientMessageId: 'stop:operation-1' } as const - await rig.journal.appendItem(op, unconfirmed, { - fence: 1, - turnScope: AGENT_JOURNAL_THREAD_SCOPE - }) - await turn('running') - const initial = await page({ direction: 'tail' }) - const before = rig.client.reduceStructuredAgentSession( - rig.client.EMPTY_STRUCTURED_AGENT_SESSION, - { - type: 'history-page', - page: initial - } - ) - const held = rig.journal.cursor() - await rig.journal.appendLifecycleBatch({ - settlementId: 're-key', - fence: 1, - mutations: [ - { kind: 'tombstone', identity: op }, - { kind: 'item', identity: noteIdentity, body: unconfirmed, turnScope: scope } - ] - }) - await turn('interrupted') - const replies = await subscribe(held) - const after = reduce(before, replies) - expect(after.items.some((item) => item.itemId === agentJournalItemKey(op))).toBe(false) - expect(render(after)).toEqual({ type: 'text', text: ordinary.text }) - }) } ) } diff --git a/tests/e2e/golden-core-flows.spec.ts b/tests/e2e/golden-core-flows.spec.ts deleted file mode 100644 index bbe65aa873d..00000000000 --- a/tests/e2e/golden-core-flows.spec.ts +++ /dev/null @@ -1,495 +0,0 @@ -import { - openSidebarProjectDialog, - openSidebarWorkspaceComposer -} from './helpers/sidebar-project-dialog' -import { execFileSync } from 'node:child_process' -import { mkdirSync, realpathSync, rmSync, writeFileSync } from 'node:fs' -import { mkdtemp } from 'node:fs/promises' -import os from 'node:os' -import path from 'node:path' -import type { ElectronApplication, Page } from '@stablyai/playwright-test' -import { test, expect } from './helpers/orca-app' -import { chooseOnboardingAgentSurface } from './helpers/onboarding-navigation' -import { ensureTerminalVisible, waitForActiveWorktree, waitForSessionReady } from './helpers/store' -import { - countVisibleTerminalPanes, - splitActiveTerminalPane, - waitForActiveTerminalManager, - waitForPaneCount, - waitForPaneIdentitySnapshot -} from './helpers/terminal' - -const tempRoots: string[] = [] -const SORTABLE_TAB = '[data-testid="sortable-tab"]' -const TASK_SOURCES_HEADING = /Set up GitHub tasks|Connect your task sources/i -const WINDOWS_TERMINAL_HEADING = /Set Windows terminal defaults/i -const ONBOARDING_ADVANCE_LABEL = /^Continue\b|^Add your first project\b/ -test.describe.configure({ mode: 'serial' }) -test.afterAll(() => { - for (const root of tempRoots.splice(0)) { - rmSync(root, { recursive: true, force: true }) - } -}) -function escapeRegExp(value: string): string { - return value.replace(/[.*+?^${}()|[\]\\]/g, '\\$&') -} - -async function createGitRepo(prefix: string, name: string): Promise { - // Resolve macOS /var symlinks like git does when registering repo.path. - const rootPath = realpathSync(await mkdtemp(path.join(os.tmpdir(), prefix))) - tempRoots.push(rootPath) - const repoPath = path.join(rootPath, name) - - mkdirSync(repoPath, { recursive: true }) - execFileSync('git', ['init'], { cwd: repoPath, stdio: 'pipe' }) - execFileSync('git', ['config', 'user.email', 'e2e@test.local'], { - cwd: repoPath, - stdio: 'pipe' - }) - execFileSync('git', ['config', 'user.name', 'E2E Test'], { - cwd: repoPath, - stdio: 'pipe' - }) - writeFileSync(path.join(repoPath, 'README.md'), `# ${name}\n`) - execFileSync('git', ['add', 'README.md'], { cwd: repoPath, stdio: 'pipe' }) - execFileSync('git', ['commit', '-m', 'Initial commit'], { cwd: repoPath, stdio: 'pipe' }) - execFileSync('git', ['branch', '-M', 'main'], { cwd: repoPath, stdio: 'pipe' }) - return repoPath -} - -async function chooseFolderInNativeDialog( - electronApp: ElectronApplication, - folderPath: string -): Promise { - for (let attempt = 0; attempt < 3; attempt += 1) { - try { - await electronApp.evaluate(({ dialog }, selectedPath) => { - dialog.showOpenDialog = async () => ({ - canceled: false, - filePaths: [selectedPath], - bookmarks: [] - }) - }, folderPath) - return - } catch (error) { - if ( - attempt === 2 || - !(error instanceof Error) || - !error.message.includes('Execution context was destroyed') - ) { - throw error - } - await new Promise((resolve) => setTimeout(resolve, 250)) - } - } -} -function onboardingFooter(page: Page) { - return page - .locator('footer') - .filter({ - has: page.getByRole('button', { name: /Back|Continue|Add your first project|Skip/i }) - }) - .first() -} - -async function continueOnboarding(page: Page): Promise { - await onboardingFooter(page).getByRole('button', { name: ONBOARDING_ADVANCE_LABEL }).click() -} - -async function selectCodexAgent(page: Page): Promise { - const codexButton = page.getByRole('button', { name: /^Codex\b/ }) - const codexVisible = await codexButton - .first() - .waitFor({ state: 'visible', timeout: 1_000 }) - .then(() => true) - .catch(() => false) - if (!codexVisible) { - await page.getByText(/Show \d+ more agents/).click() - } - await codexButton.first().click() - await expect(codexButton.first()).toHaveAttribute('aria-pressed', 'true') -} - -async function chooseOppositeTheme(page: Page): Promise { - await page.waitForFunction( - () => - document.documentElement.classList.contains('dark') || - document.documentElement.classList.contains('light') - ) - const startingTheme = await page.evaluate(() => - document.documentElement.classList.contains('dark') ? 'dark' : 'light' - ) - const nextTheme = startingTheme === 'dark' ? 'light' : 'dark' - const tileName = nextTheme === 'light' ? /Bright & crisp/ : /Easy on the eyes/ - await page.getByRole('radio', { name: tileName }).click() - await expect - .poll( - () => - page.evaluate(() => - document.documentElement.classList.contains('dark') ? 'dark' : 'light' - ), - { timeout: 5_000 } - ) - .toBe(nextTheme) -} - -async function chooseNotificationSound(page: Page): Promise { - const soundSelect = page.getByRole('combobox').first() - await expect(soundSelect).toContainText(/System Default/i) - await soundSelect.click() - const dingOption = page.getByRole('option', { name: /^Ding$/i }) - await expect(dingOption).toBeVisible() - await dingOption.press('Enter') - await expect(soundSelect).toContainText(/Ding/i) -} - -async function continueThroughOptionalSetupToNotifications(page: Page): Promise { - const taskSourcesVisible = await page - .getByRole('heading', { name: TASK_SOURCES_HEADING }) - .waitFor({ state: 'visible', timeout: 1_000 }) - .then(() => true) - .catch(() => false) - if (taskSourcesVisible) { - await continueOnboarding(page) - } - const windowsTerminalVisible = await page - .getByRole('heading', { name: WINDOWS_TERMINAL_HEADING }) - .waitFor({ state: 'visible', timeout: 1_000 }) - .then(() => true) - .catch(() => false) - if (windowsTerminalVisible) { - await continueOnboarding(page) - } - await expect(page.getByRole('heading', { name: /Set up notifications/i })).toBeVisible() -} - -async function completeOnboardingIntoHome(page: Page): Promise { - await continueOnboarding(page) - await expect(page.locator('[data-onboarding-overlay]')).toHaveCount(0) - await expect(page.getByRole('dialog', { name: /Add a project/i })).toHaveCount(0) -} - -async function waitForRepoLoaded(page: Page, repoPath: string): Promise { - await expect - .poll( - () => - page.evaluate((targetPath) => { - const state = window.__store?.getState() - const repo = state?.repos.find((candidate) => candidate.path === targetPath) - if (!state || !repo) { - return false - } - return (state.worktreesByRepo[repo.id] ?? []).length > 0 - }, repoPath), - { timeout: 30_000, message: `repo did not load: ${repoPath}` } - ) - .toBe(true) -} - -async function expectProjectVisible(page: Page, repoPath: string): Promise { - const repoName = path.basename(repoPath) - await expect(page.getByText(repoName, { exact: true }).first()).toBeVisible({ timeout: 15_000 }) -} - -async function addProjectFromSidebar( - page: Page, - electronApp: ElectronApplication, - repoPath: string -): Promise { - await chooseFolderInNativeDialog(electronApp, repoPath) - await openSidebarProjectDialog(page) - const addDialog = page.getByRole('dialog', { name: /Add a project/i }) - await expect(addDialog).toBeVisible() - await addDialog.getByRole('button', { name: /Browse folder/i }).click() - - const confirmDialog = page.getByRole('dialog', { name: /^Add Project$/i }) - const needsConfirmation = await confirmDialog - .waitFor({ state: 'visible', timeout: 2_000 }) - .then(() => true) - .catch(() => false) - if (needsConfirmation) { - await confirmDialog.getByRole('button', { name: /^Add Project$/ }).click() - } - - await waitForRepoLoaded(page, repoPath) - await expectProjectVisible(page, repoPath) -} - -async function createWorkspace(page: Page, workspaceName: string): Promise { - await openSidebarWorkspaceComposer(page) - const dialog = page.getByRole('dialog', { name: /Create (Workspace|Worktree)/i }) - await expect(dialog).toBeVisible() - const nameInput = dialog.getByPlaceholder(/Type a name/i) - await expect(nameInput).toBeVisible() - await nameInput.fill(workspaceName) - await dialog.getByRole('button', { name: /Create (Workspace|Worktree)/i }).click() - await expect(dialog).toBeHidden({ timeout: 20_000 }) - await expectActiveWorkspaceVisible(page, workspaceName) -} - -async function expectActiveWorkspaceBelongsToRepo( - page: Page, - workspaceName: string, - repoPath: string -): Promise { - await expectActiveWorkspaceVisible(page, workspaceName) - await expect - .poll( - () => - page.evaluate( - ({ targetRepoPath, targetWorkspaceName }) => { - const state = window.__store?.getState() - if (!state?.activeWorktreeId) { - return null - } - const repo = state.repos.find((candidate) => candidate.path === targetRepoPath) - if (!repo) { - return null - } - const activeWorktree = (state.worktreesByRepo[repo.id] ?? []).find( - (worktree) => worktree.id === state.activeWorktreeId - ) - return activeWorktree?.displayName === targetWorkspaceName - }, - { targetRepoPath: repoPath, targetWorkspaceName: workspaceName } - ), - { timeout: 20_000, message: 'active workspace did not belong to the newly added project' } - ) - .toBe(true) -} - -async function expectActiveWorkspaceVisible(page: Page, workspaceName: string): Promise { - const activeWorkspace = page - .locator('[role="option"][aria-current="page"]') - .filter({ hasText: new RegExp(escapeRegExp(workspaceName)) }) - .first() - await expect(activeWorkspace).toBeVisible({ timeout: 20_000 }) -} - -async function countRenderedTabs(page: Page): Promise { - return page.locator(SORTABLE_TAB).count() -} - -async function renderedTabIds(page: Page): Promise { - return page.locator(SORTABLE_TAB).evaluateAll((tabs) => - tabs.flatMap((tab) => { - const tabId = tab.getAttribute('data-tab-id') - return tabId ? [tabId] : [] - }) - ) -} - -async function expectTerminalSurface(page: Page): Promise { - const terminalSurface = page - .locator('[data-terminal-tab-id][data-terminal-layout-leaf-ids]:visible') - .first() - await expect(terminalSurface).toBeVisible({ timeout: 30_000 }) - await expect(terminalSurface).not.toHaveAttribute('data-os-file-drop-owner') - await expect(terminalSurface.locator('.pane[data-leaf-id]').first()).toHaveAttribute( - 'data-os-file-drop-owner', - '' - ) -} - -async function waitForTerminalPaneManager(page: Page): Promise { - await waitForPaneCount(page, 1, 30_000) - await waitForActiveTerminalManager(page, 30_000) -} - -async function createTerminalTabThroughMenu(page: Page): Promise { - const tabIdsBefore = await renderedTabIds(page) - await page.getByRole('button', { name: 'New tab' }).click({ force: true }) - const newTerminalMenuItem = page.getByRole('menuitem', { name: /New Terminal/i }).first() - await newTerminalMenuItem.click({ force: true }) - await expect.poll(() => countRenderedTabs(page), { timeout: 5_000 }).toBe(tabIdsBefore.length + 1) - const createdTabIds = (await renderedTabIds(page)).filter( - (tabId) => !tabIdsBefore.includes(tabId) - ) - expect(createdTabIds, 'new terminal tab should render exactly one new tab').toHaveLength(1) - const createdTab = page.locator(`${SORTABLE_TAB}[data-tab-id="${createdTabIds[0]}"]`).first() - await expect(createdTab).toHaveAttribute('data-tab-title', /.+/) - await expectTerminalSurface(page) -} - -async function splitTerminalPaneAndAssertIdentity(page: Page): Promise { - const paneCountBefore = await countVisibleTerminalPanes(page) - await splitActiveTerminalPane(page, 'vertical') - await waitForPaneCount(page, paneCountBefore + 1) - const snapshot = await waitForPaneIdentitySnapshot(page, paneCountBefore + 1) - expect(snapshot.panes).toHaveLength(paneCountBefore + 1) -} - -async function requestAgentSessionsTour(page: Page): Promise { - await expect - .poll( - () => - page.evaluate(() => { - const state = window.__store?.getState() - const splitTarget = document.querySelector( - '[data-contextual-tour-target="terminal-pane-split-target"], [data-contextual-tour-target="workspace-agent-terminal-tip"]' - ) - const rect = splitTarget?.getBoundingClientRect() - return { - ready: state?.persistedUIReady === true, - onboardingHidden: state?.contextualToursOnboardingVisible === false, - noModal: state?.activeModal === 'none', - splitTargetMeasurable: Boolean(rect && rect.width > 0 && rect.height > 0) - } - }), - { timeout: 30_000 } - ) - .toEqual({ - ready: true, - onboardingHidden: true, - noModal: true, - splitTargetMeasurable: true - }) - - await page.evaluate(() => { - window.__store - ?.getState() - .requestContextualTour('workspace-agent-sessions', 'setup_guide_parallel_work', false, { - force: true - }) - }) - const newTabStep = page.getByRole('dialog', { name: 'New agents start here', exact: true }) - await expect(newTabStep).toBeVisible() - await page - .locator( - '[data-workspace-surface-id][aria-hidden="false"] [data-contextual-tour-target="tab-create-menu"]' - ) - .click() - await expect(page.getByRole('menu')).toBeVisible() - await expect(newTabStep).toBeHidden() - await expect(page.getByRole('dialog', { name: /Split a terminal pane/i })).toBeHidden() - await expect - .poll(() => page.evaluate(() => window.__store?.getState().activeContextualTourStepIndex)) - .toBe(0) - await page.keyboard.press('Escape') - await expect(page.getByRole('menu')).toBeHidden() - const splitStep = page.getByRole('dialog', { name: /Split a terminal pane/i }) - await expect(splitStep).toBeVisible() - await splitStep.getByRole('button', { name: 'Back', exact: true }).click() - await expect(newTabStep).toBeVisible() - await page - .locator( - '[data-workspace-surface-id][aria-hidden="false"] [data-contextual-tour-target="tab-create-menu"]' - ) - .click() - await expect(newTabStep).toBeHidden() - await page.getByRole('menuitem', { name: /^New Terminal/ }).click() - await expect(page.getByRole('menu')).toBeHidden() - await expect(page.getByRole('dialog', { name: /Split a terminal pane/i })).toBeVisible() -} - -async function completeWorkspaceCreationTour(page: Page, workspaceName: string): Promise { - const pickProjectStep = page.getByRole('dialog', { name: /Pick a project/i }) - await expect(pickProjectStep).toBeVisible() - // Keyboard-activate Next so a lingering project popover cannot obscure the tour transition. - await pickProjectStep.getByRole('button', { name: /^Next$/ }).focus() - await page.keyboard.press('Enter') - const nameStep = page.getByRole('dialog', { name: /Name it, or start from existing work/i }) - await expect(nameStep).toBeVisible() - const autoNameSwitch = nameStep.getByRole('switch', { - name: /Auto-name workspace from first agent message/i - }) - const checkedBefore = await autoNameSwitch.getAttribute('aria-checked') - await autoNameSwitch.click() - await expect(autoNameSwitch).toHaveAttribute( - 'aria-checked', - checkedBefore === 'true' ? 'false' : 'true' - ) - await page.getByRole('button', { name: /^Next$/ }).click() - await expect( - page.getByRole('dialog', { name: /Choose what agent starts the work/i }) - ).toBeVisible() - await page.getByRole('button', { name: /^Done$/ }).click() - - const composer = page.getByRole('dialog', { name: /Create (Workspace|Worktree)/i }) - await expect(composer).toBeVisible() - await composer.getByPlaceholder(/Type a name/i).fill(workspaceName) - await composer.getByRole('button', { name: /Create (Workspace|Worktree)/i }).click() - await expect(composer).toBeHidden({ timeout: 20_000 }) - await expectActiveWorkspaceVisible(page, workspaceName) -} - -test.describe('Existing-user golden core flow', () => { - test('adds project, creates workspace, opens a terminal tab, and splits a pane', async ({ - electronApp, - orcaPage - }) => { - await waitForSessionReady(orcaPage) - await waitForActiveWorktree(orcaPage) - const repoPath = await createGitRepo('orca-e2e-golden-existing-', 'golden-existing-project') - - await addProjectFromSidebar(orcaPage, electronApp, repoPath) - const workspaceName = `golden-existing-${Date.now()}` - await createWorkspace(orcaPage, workspaceName) - await expectActiveWorkspaceBelongsToRepo(orcaPage, workspaceName, repoPath) - await ensureTerminalVisible(orcaPage) - await expectTerminalSurface(orcaPage) - await waitForTerminalPaneManager(orcaPage) - - await createTerminalTabThroughMenu(orcaPage) - await splitTerminalPaneAndAssertIdentity(orcaPage) - }) -}) - -test.describe('New-user golden core flow', () => { - test.use({ dismissOnboarding: false, seedTestRepo: false }) - - test('completes onboarding, adds a project, and follows the workspace tour handoff', async ({ - electronApp, - orcaPage - }) => { - await waitForSessionReady(orcaPage) - await chooseOnboardingAgentSurface(orcaPage) - - await selectCodexAgent(orcaPage) - await continueOnboarding(orcaPage) - await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible() - await chooseOppositeTheme(orcaPage) - await continueOnboarding(orcaPage) - await continueThroughOptionalSetupToNotifications(orcaPage) - await expect(orcaPage.getByRole('button', { name: /Send Test Notification/i })).toBeVisible() - await chooseNotificationSound(orcaPage) - await completeOnboardingIntoHome(orcaPage) - await orcaPage.getByRole('button', { name: 'Add project', exact: true }).click() - await expect(orcaPage.getByRole('dialog', { name: /Add a project/i })).toBeVisible() - - const repoPath = await createGitRepo('orca-e2e-golden-new-', 'golden-new-project') - await chooseFolderInNativeDialog(electronApp, repoPath) - await orcaPage - .getByRole('button', { name: /Browse for a folder|Open a folder|Browse folder/i }) - .click() - await waitForRepoLoaded(orcaPage, repoPath) - await expectProjectVisible(orcaPage, repoPath) - await waitForActiveWorktree(orcaPage) - await ensureTerminalVisible(orcaPage) - await expectTerminalSurface(orcaPage) - await waitForTerminalPaneManager(orcaPage) - - await requestAgentSessionsTour(orcaPage) - const paneCountBeforeTourSplit = await countVisibleTerminalPanes(orcaPage) - await orcaPage.getByRole('button', { name: /^Split terminal$/ }).click() - await waitForPaneCount(orcaPage, paneCountBeforeTourSplit + 1) - await waitForPaneIdentitySnapshot(orcaPage, paneCountBeforeTourSplit + 1) - - await expect( - orcaPage.getByRole('dialog', { name: /Start another task in parallel/i }) - ).toBeVisible() - const createControl = orcaPage - .locator('[data-contextual-tour-target="workspace-create-control"]') - .first() - await expect(createControl).toBeVisible() - await expect(createControl).toHaveAttribute('aria-label', 'New workspace') - const createControlBox = await createControl.boundingBox() - expect(createControlBox?.width ?? 0).toBeGreaterThan(0) - expect(createControlBox?.height ?? 0).toBeGreaterThan(0) - await createControl.click() - - const workspaceName = `golden-new-${Date.now()}` - await completeWorkspaceCreationTour(orcaPage, workspaceName) - await expectActiveWorkspaceBelongsToRepo(orcaPage, workspaceName, repoPath) - }) -}) diff --git a/tests/e2e/helpers/orcad-convert-teardown.ts b/tests/e2e/helpers/orcad-convert-teardown.ts deleted file mode 100644 index 1c26c4507fb..00000000000 --- a/tests/e2e/helpers/orcad-convert-teardown.ts +++ /dev/null @@ -1,57 +0,0 @@ -import type { OrcadManagedStopResult } from '../../../src/shared/orcad-managed-runtime' - -export type ConvertedHostTeardown = { - environmentId: string - targetId: string - worktreeId: string -} - -/** Runs in the renderer; only this fixture's converted SSH environment can be stopped. */ -export async function decommissionConvertedHost({ - environmentId, - targetId, - worktreeId -}: ConvertedHostTeardown): Promise { - const managedOrcad = window.api.runtimeEnvironments.managedOrcad - if (!managedOrcad) { - throw new Error('Converted host teardown requires the desktop managed server API') - } - const environments = await window.api.runtimeEnvironments.list() - const owned = environments.filter( - (environment) => - environment.id === environmentId && - environment.orcadDeployment?.kind !== 'wsl' && - environment.orcadDeployment?.sshTargetId === targetId - ) - if (owned.length !== 1) { - throw new Error('Converted host teardown could not verify its environment and SSH target') - } - const closed = await window.api.runtimeEnvironments.call({ - selector: environmentId, - method: 'terminal.closeAll', - params: { worktree: `id:${worktreeId}` }, - timeoutMs: 120_000 - }) - const receipt = closed.ok ? closed.result : null - if ( - !receipt || - typeof receipt !== 'object' || - !('retiredSurfaces' in receipt) || - receipt.retiredSurfaces !== true || - ('ptyStopVerdict' in receipt && receipt.ptyStopVerdict !== undefined) - ) { - throw new Error(`Converted host terminal stop was not proven: ${JSON.stringify(closed)}`) - } - const stopped = await managedOrcad.stop({ - selector: environmentId - }) - if ( - stopped.outcome !== 'unlinked' || - stopped.verdict !== 'exited' || - stopped.environmentId !== environmentId || - stopped.sshTargetId !== targetId - ) { - throw new Error(`Converted host decommission was not proven: ${JSON.stringify(stopped)}`) - } - return stopped -} diff --git a/tests/e2e/helpers/windows-host-orcad-processes.ts b/tests/e2e/helpers/windows-host-orcad-processes.ts index e0ac530194a..3ed86cf841a 100644 --- a/tests/e2e/helpers/windows-host-orcad-processes.ts +++ b/tests/e2e/helpers/windows-host-orcad-processes.ts @@ -51,7 +51,7 @@ export async function listHostOrcadProcesses(home: string): Promise { return (await listHostOrcadProcesses(home)).filter((entry) => - /[\\/]orcad\.js(?=["\s]|$)/iu.test(entry.commandLine) + /[\\/]orcad(?:-server)?\.js(?=["\s]|$)/iu.test(entry.commandLine) ) } diff --git a/tests/e2e/ssh-orcad-auto-convert.spec.ts b/tests/e2e/ssh-orcad-auto-convert.spec.ts deleted file mode 100644 index 770387ee214..00000000000 --- a/tests/e2e/ssh-orcad-auto-convert.spec.ts +++ /dev/null @@ -1,488 +0,0 @@ -// #24979: convert real relay-era repo/folder/editor state on first SSH connect, retaining hidden source rows. -// Also update bundled orcad on idle reconnect and handle an occupied preferred port. -// ORCA_E2E_ORCAD_CONVERT_HOST selects Docker or a Windows host-cell descriptor; ORCA_E2E_ORCAD_CONVERT_TEMPLATE must match its target. -import { cpSync, mkdirSync, mkdtempSync, rmSync } from 'node:fs' -import os from 'node:os' -import path from 'node:path' -import type { ElectronApplication, Page, TestInfo } from '@stablyai/playwright-test' -import { expect, test } from './helpers/orca-app' -import { waitForSessionReady } from './helpers/store' -import { createRestartSession } from './helpers/orca-restart' -import { - convertAndRetain, - managedServer, - reconnect, - serverCall -} from './helpers/orcad-convert-flow' -import { - isOrcadFullVersion, - makeOrcadTemplateVariant, - readHostOrcadActivation -} from './helpers/orcad-template-variant' -import { seedRelayEraProfile } from './helpers/orcad-upgrade-profile' -import { findOrcadMigrationSourceCutoverForTarget } from '../../src/main/persistence/migrating-orcad-catalog/cutover-journals/orcad-migration-cutover-journal' -import { toSshExecutionHostId } from '../../src/shared/execution-host' -import { - decommissionConvertedHost, - type ConvertedHostTeardown -} from './helpers/orcad-convert-teardown' -import { - ORCAD_CONVERT_HOST_ENV, - startOrcadConvertHost, - type OrcadConvertHost -} from './helpers/orcad-convert-host' -import { - callEnvironment, - createPairedHostTerminal, - openPairedClientTab -} from './helpers/paired-host-terminal' -import { expectTerminalAccessibilityText } from './helpers/terminal-accessibility-tree' -import { getOrcadSshDeployment } from '../../src/shared/remote-runtime/host-catalog/runtime-environments' - -const HOST = process.env[ORCAD_CONVERT_HOST_ENV] -const TEMPLATE_SOURCE = process.env.ORCA_E2E_ORCAD_CONVERT_TEMPLATE -// Fixed per worker so the app's launch env can name them before the test runs. -const SCRATCH = path.join(os.tmpdir(), `orca-orcad-convert-${process.pid}`) -const TEMPLATE_DIR = path.join(SCRATCH, 'orcad-template') - -// Whole file: every test needs a real host and a template built for it. -test.skip( - !HOST || !TEMPLATE_SOURCE, - `Set ${ORCAD_CONVERT_HOST_ENV} and ORCA_E2E_ORCAD_CONVERT_TEMPLATE` -) - -let host: OrcadConvertHost | null = null -function startHost(testInfo: TestInfo): OrcadConvertHost { - host = startOrcadConvertHost(HOST!, testInfo) - return host -} - -test.beforeEach(() => { - rmSync(SCRATCH, { recursive: true, force: true }) - mkdirSync(SCRATCH, { recursive: true }) -}) - -test.afterEach(() => { - host?.cleanup() - host = null - rmSync(SCRATCH, { recursive: true, force: true }) -}) - -test.use({ - orcaAppExtraEnv: { - ORCA_ORCAD_TEMPLATE_PATH: TEMPLATE_DIR - } -}) - -test('a relay-era profile converts its SSH host on the first connect after upgrading', async ({ - testRepoPath -}, testInfo) => { - test.setTimeout(20 * 60_000) - const host = startHost(testInfo) - const session = createRestartSession(testInfo, { - ORCA_ORCAD_TEMPLATE_PATH: TEMPLATE_SOURCE! - }) - let app: ElectronApplication | null = null - let teardown: (ConvertedHostTeardown & { page: Page }) | null = null - try { - // The first launch only establishes the profile the relay-era rows are written into. - const first = await session.launch() - app = first.app - await waitForSessionReady(first.page) - await first.page.evaluate((repoPath) => window.api.repos.add({ path: repoPath }), testRepoPath) - await session.close(app) - app = null - const seeded = seedRelayEraProfile( - session.userDataDir, - host.input, - { repoPath: host.remoteRepoPath, folderPath: host.remoteFolderPath }, - { terminalTab: true, parkedScrollback: true } - ) - // With no relay leases, the gate must prove exited; retain raw SSH logs because this launch bypasses fixture forwarding. - const upgraded = await session.launch({ - onStderr: (chunk) => { - if (chunk.includes('[ssh]')) { - process.stderr.write(chunk) - } - } - }) - app = upgraded.app - await waitForSessionReady(upgraded.page) - // Mounting the unconverted relay tab dials the host and triggers conversion without spawning locally. - await upgraded.page.evaluate( - ({ worktreeId, hostId, tabId }) => { - const state = window.__store!.getState() - state.setActiveWorktree(worktreeId, hostId) - state.setActiveTab(tabId) - state.setActiveTabType('terminal', worktreeId) - }, - { - worktreeId: seeded.worktreeId, - hostId: toSshExecutionHostId(seeded.targetId), - tabId: seeded.terminalTabId! - } - ) - await expect - .poll(() => managedServer(upgraded.page, seeded.targetId), { timeout: 8 * 60_000 }) - .toMatchObject({ kind: 'managed' }) - await upgraded.page.screenshot({ path: testInfo.outputPath('relay-era-terminal-dialed.png') }) - await convertAndRetain(upgraded.page, session.userDataDir, seeded) - // #30126: the parked panes' history rode the snapshot channel; inlined, it overflowed the manifest. - const moved = findOrcadMigrationSourceCutoverForTarget(session.userDataDir, seeded.targetId) - ?.manifest.payload.dormantState - expect(moved?.workspaceSession?.localOnlyScrollbackByTabId).toBeUndefined() - expect( - Object.fromEntries( - (moved?.terminalScrollbackSnapshots ?? []) - .filter((snapshot) => snapshot.tabId === seeded.terminalTabId) - .map((snapshot) => [snapshot.leafId, snapshot.byteLength]) - ) - ).toEqual( - Object.fromEntries( - Object.entries(seeded.parkedScrollbackByLeafId).map(([leafId, buffer]) => [ - leafId, - Buffer.byteLength(buffer) - ]) - ) - ) - const environment = ( - await upgraded.page.evaluate(() => window.api.runtimeEnvironments.list()) - ).find((entry) => getOrcadSshDeployment(entry)?.sshTargetId === seeded.targetId) - if (!environment) { - throw new Error('Converted host is missing from the environment catalog') - } - teardown = { - page: upgraded.page, - environmentId: environment.id, - targetId: seeded.targetId, - worktreeId: seeded.worktreeId - } - const hostId = `runtime:${environment.id}` as const - await upgraded.page.evaluate((hostId) => { - const state = window.__store!.getState() - state.setGroupBy('none') - window.__store!.setState({ visibleWorkspaceHostIds: ['local', hostId] }) - }, hostId) - const folder = upgraded.page.getByText('orcad upgrade folder', { exact: true }) - await expect(folder).toBeVisible() - await upgraded.page.screenshot({ path: testInfo.outputPath('managed-folder-host-section.png') }) - const sectionHeader = folder.locator('xpath=preceding::*[@data-host-header-drag-id][1]') - await expect(sectionHeader).toHaveAttribute('data-host-header-drag-id', hostId) - const managedHeader = upgraded.page.locator(`[data-host-header-drag-id="${hostId}"]`) - await managedHeader.click() - await expect(folder).toBeHidden() - await managedHeader.click() - await expect(folder).toBeVisible() - if (!host.exec) { - return - } - await upgraded.page.evaluate( - ({ worktreeId, environmentId }) => { - window.__store!.getState().setActiveWorktree(worktreeId, `runtime:${environmentId}`) - }, - { worktreeId: seeded.worktreeId, environmentId: environment.id } - ) - const terminal = await createPairedHostTerminal( - upgraded.page, - environment.id, - seeded.worktreeId, - 'bash' - ) - await openPairedClientTab(upgraded.page, seeded.worktreeId, terminal.webTabId) - await callEnvironment(upgraded.page, environment.id, 'terminal.send', { - terminal: terminal.terminal, - text: "orca status --json > /tmp/orca-cli-status.json && orca worktree ps --json > /tmp/orca-cli-workers.json && printf 'QA_%s\\n' 'CLI_READY'", - enter: true - }) - await expectTerminalAccessibilityText(upgraded.page, terminal.webTabId, 'QA_CLI_READY') - const status = JSON.parse(host.exec('cat /tmp/orca-cli-status.json')) - expect(status.result.runtime).toMatchObject({ - reachable: true, - runtimeId: environment.runtimeId - }) - expect(status.result.app.desktopWindowStatus).toBe('blocked') - await upgraded.page.screenshot({ path: testInfo.outputPath('managed-cli-ready.png') }) - } finally { - try { - // The Windows account survives client exit; stop its owned runtime before the lane's grace. - if (HOST !== 'docker' && teardown) { - const { page, ...owned } = teardown - const stopped = await page.evaluate(decommissionConvertedHost, owned) - await testInfo.attach('converted-host-teardown', { - body: JSON.stringify(stopped), - contentType: 'application/json' - }) - } - } finally { - try { - if (app) { - await session.close(app) - } - } finally { - await session.dispose() - } - } - } -}) - -test('a host whose sshd refuses TCP forwarding runs a managed server over the stdio bridge', async ({ - orcaPage: page -}, testInfo) => { - test.skip(HOST !== 'docker', 'Only the Docker host can change its sshd policy mid-test') - test.setTimeout(15 * 60_000) - cpSync(TEMPLATE_SOURCE!, TEMPLATE_DIR, { recursive: true }) - const host = startHost(testInfo) - host.blockTcpForwarding!() - await waitForSessionReady(page) - // Just the connect: a managed host serves repositories through its server, not the relay. - const targetId = await page.evaluate(async (input) => { - const { target } = await window.api.ssh.addTarget({ target: input }) - await window.api.ssh.connect({ targetId: target.id }) - return target.id - }, host.input) - const mainServer = (): Promise => - page.evaluate( - async (id) => - JSON.stringify((await window.api.ssh.getState({ targetId: id }))?.managedServer ?? null), - targetId - ) - - await expect.poll(mainServer, { timeout: 8 * 60_000 }).toContain('"kind":"managed"') - const target = await page.evaluate( - async (id) => (await window.api.ssh.listTargets()).find((entry) => entry.id === id), - targetId - ) - expect(target?.orcadFence).toBeTruthy() - expect(target).not.toHaveProperty('managedServerUnavailable') - const environment = (await page.evaluate(() => window.api.runtimeEnvironments.list())).find( - (entry) => getOrcadSshDeployment(entry)?.sshTargetId === targetId - ) - expect(environment, 'a managed server registered for the host').toBeTruthy() - // sshd refuses every forward, so this call can only have ridden a stdio bridge. - await serverCall(page, environment!.id, 'repo.list') - - // A reconnect rebuilds the tunnel the same way. - await reconnect(page, targetId) - await expect.poll(mainServer, { timeout: 2 * 60_000 }).toContain('"kind":"managed"') - await serverCall(page, environment!.id, 'repo.list') -}) - -const heldPortProbe = `(echo > /dev/tcp/127.0.0.1/6768) 2>/dev/null && echo held || true` -// Every installed slot's readiness line names the endpoint that orcad actually bound. -const BOUND_ENDPOINTS = `cat /root/.orca-remote/orcad-*/.orcad-readiness 2>/dev/null | grep -o 'ws://[0-9.:]*' || true` - -test('a host whose port 6768 another runtime holds still runs a managed server', async ({ - orcaPage: page -}, testInfo) => { - test.skip(HOST !== 'docker', 'Only the Docker host can start a listener mid-test') - test.setTimeout(15 * 60_000) - cpSync(TEMPLATE_SOURCE!, TEMPLATE_DIR, { recursive: true }) - const host = startHost(testInfo) - // Stands in for a desktop Orca on the host: it holds 6768 and turns every client away. - host.exec!( - `setsid nohup node -e "require('net').createServer((s) => s.destroy()).listen(6768, '127.0.0.1')" >/dev/null 2>&1 &` - ) - await expect.poll(() => host.exec!(heldPortProbe)).toContain('held') - await waitForSessionReady(page) - const targetId = await page.evaluate(async (input) => { - const { target } = await window.api.ssh.addTarget({ target: input }) - await window.api.ssh.connect({ targetId: target.id }) - return target.id - }, host.input) - - await expect - .poll(() => managedServer(page, targetId), { timeout: 8 * 60_000 }) - .toMatchObject({ kind: 'managed' }) - const environment = (await page.evaluate(() => window.api.runtimeEnvironments.list())).find( - (entry) => getOrcadSshDeployment(entry)?.sshTargetId === targetId - ) - expect(environment, 'a managed server registered for the host').toBeTruthy() - // The other listener drops every connection, so this call can only have reached orcad. - await serverCall(page, environment!.id, 'repo.list') - // Without these the case passes vacuously whenever the stand-in is gone before orcad starts. - expect(host.exec!(heldPortProbe)).toContain('held') - expect(host.exec!(BOUND_ENDPOINTS)).toMatch(/ws:\/\/127\.0\.0\.1:(?!6768\b)\d+/) - - // A reconnect reads the bound port again rather than assuming 6768. - await reconnect(page, targetId) - await expect - .poll(() => managedServer(page, targetId), { timeout: 2 * 60_000 }) - .toMatchObject({ kind: 'managed' }) - await serverCall(page, environment!.id, 'repo.list') -}) - -test('a managed host updates to the bundled orcad on the first connect after an app update', async (// oxlint-disable-next-line no-empty-pattern -- Playwright's second fixture arg is testInfo; the first must be an object destructure to opt out of the default fixture set. -{}, testInfo) => { - test.skip(HOST !== 'docker', 'Reads the activation record through the Docker host') - test.setTimeout(20 * 60_000) - cpSync(TEMPLATE_SOURCE!, TEMPLATE_DIR, { recursive: true }) - const host = startHost(testInfo) - const session = createRestartSession(testInfo, { - ORCA_ORCAD_TEMPLATE_PATH: TEMPLATE_DIR - }) - let app: ElectronApplication | null = null - try { - const first = await session.launch() - app = first.app - await waitForSessionReady(first.page) - const targetId = await first.page.evaluate(async (input) => { - const { target } = await window.api.ssh.addTarget({ target: input }) - return target.id - }, host.input) - await connectOnce(first.page, targetId) - await expect - .poll(() => managedServer(first.page, targetId), { timeout: 8 * 60_000 }) - .toMatchObject({ kind: 'managed' }) - const deployed = readHostOrcadActivation(host.exec!) - expect(isOrcadFullVersion(deployed.active)).toBe(true) - await session.close(app) - app = null - - // Use template B to model a bundled app update during tunnel restore/next connect. - makeOrcadTemplateVariant(TEMPLATE_DIR, 'B') - const updated = await session.launch({ - onStderr: (chunk) => { - if (chunk.includes('[ssh]')) { - process.stderr.write(chunk) - } - } - }) - app = updated.app - await waitForSessionReady(updated.page) - // Connect directly: a disconnect racing launch-time tunnel restore cancels the update-triggering connect. - const attempts: string[] = [] - await expect - .poll( - async () => { - attempts.push(await connectOnce(updated.page, targetId)) - const record = readHostOrcadActivation(host.exec!) - return record.active !== deployed.active - ? 'updated' - : JSON.stringify({ attempts: attempts.slice(-3), record }) - }, - { timeout: 8 * 60_000, intervals: [5_000] } - ) - .toBe('updated') - await expect - .poll(() => managedServer(updated.page, targetId), { timeout: 60_000 }) - .toEqual({ kind: 'managed', environmentId: expect.any(String) }) - const record = readHostOrcadActivation(host.exec!) - expect(isOrcadFullVersion(record.active)).toBe(true) - expect(record.previous).toBe(deployed.active) - expect(record.activeAppVersion).toBeTruthy() - } finally { - if (app) { - await session.close(app) - } - await session.dispose() - } -}) - -/** One connect; returns its managed-server state, or why it threw, for a poll to report. */ -function connectOnce(page: Page, targetId: string): Promise { - return page.evaluate(async (id) => { - try { - const state = await window.api.ssh.connect({ targetId: id }) - return JSON.stringify(state?.managedServer ?? null) - } catch (error) { - return `connect threw: ${String(error)}` - } - }, targetId) -} - -test('reconnect restores the managed host name after its conversion catalog fails', async ({ - testRepoPath -}, testInfo) => { - test.skip(HOST !== 'docker', 'Catalog fault injection uses the isolated Linux Docker host') - test.setTimeout(180_000) - const scratch = mkdtempSync(path.join(os.tmpdir(), 'orca-catalog-retry-')) - const templateDir = path.join(scratch, 'template') - const host = startOrcadConvertHost('docker', testInfo) - const session = createRestartSession(testInfo, { ORCA_ORCAD_TEMPLATE_PATH: templateDir }) - let app: Awaited>['app'] | null = null - try { - const first = await session.launch() - app = first.app - await waitForSessionReady(first.page) - // A local project keeps a second host section, so the managed host header renders. - await first.page.evaluate((repoPath) => window.api.repos.add({ path: repoPath }), testRepoPath) - await session.close(app) - app = null - const seeded = seedRelayEraProfile(session.userDataDir, host.input, { - repoPath: host.remoteRepoPath, - folderPath: host.remoteFolderPath - }) - // The saved profile is the relay era: no connect has run, so the first one converts. - const upgraded = await session.launch() - app = upgraded.app - const page = upgraded.page - await waitForSessionReady(page) - await page.evaluate(() => window.__store!.getState().setGroupBy('none')) - - // Fail the local catalog IPC during a real SSH migration; healing restores its production handler. - await app.evaluate(({ ipcMain }) => { - type Handler = (event: unknown, ...args: unknown[]) => unknown - // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: Electron's registered invoke handlers are a Map of channel callbacks. - const registry = (ipcMain as unknown as { _invokeHandlers?: Map }) - ._invokeHandlers - const original = registry?.get('runtimeEnvironments:list') - if (!original) { - throw new Error('runtimeEnvironments:list is not registered') - } - ipcMain.removeHandler('runtimeEnvironments:list') - ipcMain.handle('runtimeEnvironments:list', () => { - throw new Error('catalog-retry: managed-host catalog temporarily unavailable') - }) - // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: this test installs and removes its own recovery callback within this app. - const state = globalThis as typeof globalThis & { healCatalogFault?: () => void } - state.healCatalogFault = () => { - ipcMain.removeHandler('runtimeEnvironments:list') - ipcMain.handle('runtimeEnvironments:list', original) - } - }) - // Collect rather than waitForEvent: its 30 s clock would start before the conversion below. - let catalogFailed = false - page.on('console', (message) => { - catalogFailed ||= message.text().includes('Could not load the managed server catalogs') - }) - cpSync(TEMPLATE_SOURCE!, templateDir, { recursive: true }) - await reconnect(page, seeded.targetId) - await expect.poll(() => managedServer(page, seeded.targetId)).toMatchObject({ kind: 'managed' }) - await expect.poll(() => catalogFailed).toBe(true) - await page.screenshot({ path: testInfo.outputPath('catalog-failure.png') }) - - await app.evaluate(() => { - // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: the fault installation above owns this optional callback. - const state = globalThis as typeof globalThis & { healCatalogFault?: () => void } - state.healCatalogFault?.() - }) - const environment = (await page.evaluate(() => window.api.runtimeEnvironments.list())).find( - (entry) => getOrcadSshDeployment(entry)?.sshTargetId === seeded.targetId - ) - if (!environment) { - throw new Error('Managed host missing from main catalog') - } - await page.evaluate( - (id) => window.__store!.setState({ visibleWorkspaceHostIds: ['local', `runtime:${id}`] }), - environment.id - ) - await reconnect(page, seeded.targetId) - // The folder row must retain its managed execution host after the catalog heals. - const folder = page.locator('[data-worktree-host-identity]').filter({ - has: page.getByText('orcad upgrade folder', { exact: true }) - }) - await expect(folder).toHaveCount(1, { timeout: 30_000 }) - await expect(folder).toHaveAttribute('data-worktree-host-identity', /runtime:/) - await expect( - page.locator('[data-host-header-drag-id^="runtime:"]').filter({ hasText: environment.name }) - ).toHaveCount(1) - await page.screenshot({ path: testInfo.outputPath('catalog-recovered.png') }) - } finally { - if (app) { - await session.close(app) - } - await session.dispose() - host.cleanup() - rmSync(scratch, { recursive: true, force: true }) - } -}) diff --git a/tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts b/tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts index d4806971d0d..5a3ef03d45f 100644 --- a/tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts +++ b/tests/e2e/ssh-orcad-windows-cli-matrix.spec.ts @@ -1,5 +1,5 @@ -// Run bundled CLI against real Windows OpenSSH host cells; invoke-windows-host-cells.ps1 selects @orcad-cli-managed/@orcad-cli-convert. -// Managed covers deploy/list, terminal reconnect, decommission/redeploy and unverifiable→live restart; convert preserves relay-era source and reaches migrated projects. +// Run bundled CLI against real Windows OpenSSH host cells; invoke-windows-host-cells.ps1 selects @orcad-cli-managed. +// Covers deploy/list, terminal reconnect, decommission/redeploy and unverifiable→live restart. // ORCA_E2E_ORCAD_CONVERT_HOST names the descriptor; ORCA_E2E_ORCAD_CONVERT_TEMPLATE must match its target. import { mkdirSync, rmSync } from 'node:fs' import os from 'node:os' @@ -8,8 +8,8 @@ import type { ElectronApplication, Page } from '@stablyai/playwright-test' import { expect, test } from './helpers/orca-app' import { waitForSessionReady } from './helpers/store' import { createRestartSession } from './helpers/orca-restart' -import { convertAndRetain, managedServer, reconnect } from './helpers/orcad-convert-flow' -import { seedRelayEraProfile } from './helpers/orcad-upgrade-profile' +import { managedServer, reconnect } from './helpers/orcad-convert-flow' + import { ORCAD_CONVERT_HOST_ENV, startOrcadConvertHost } from './helpers/orcad-convert-host' import { orcaCliResult, runCompiledOrcaCli } from './helpers/compiled-orca-cli' import { @@ -523,51 +523,3 @@ test('@orcad-cli-managed an empty Windows host deploys, serves CLI terminals thr rmSync(SCRATCH, { recursive: true, force: true }) } }) - -test('@orcad-cli-convert a seeded relay-era profile converts its Windows host, keeps its source retained, and the CLI reaches its projects', async (// oxlint-disable-next-line no-empty-pattern -- This test owns its launches through a restart session. -{}, testInfo) => { - skipUnlessWindowsHost() - test.setTimeout(20 * 60_000) - resetScratch() - const host = startOrcadConvertHost(HOST!, testInfo) - const session = createRestartSession(testInfo, launchEnv()) - let app: ElectronApplication | null = null - try { - const first = await session.launch() - app = first.app - await waitForSessionReady(first.page) - await session.close(app) - app = null - const seeded = seedRelayEraProfile(session.userDataDir, host.input, { - repoPath: host.remoteRepoPath, - folderPath: host.remoteFolderPath - }) - const upgraded = await session.launch({ onStderr: sshStderr }) - app = upgraded.app - await waitForSessionReady(upgraded.page) - await convertAndRetain(upgraded.page, session.userDataDir, seeded) - - const environmentId = await waitForManaged(upgraded.page, seeded.targetId) - expect(await environmentIds(session.userDataDir)).toContain(environmentId) - expect( - (await hostRows(session.userDataDir)).find((row) => row.id === seeded.targetId) - ).toMatchObject({ - kind: 'ssh', - connected: true - }) - const repos = await orcaCliResult<{ repos: { path: string }[] }>(session.userDataDir, [ - 'repo', - 'list', - '--environment', - environmentId - ]) - expect(repos.repos.some((repo) => samePath(repo.path, seeded.repoPath))).toBe(true) - } finally { - if (app) { - await session.close(app) - } - await disposeQuietly(session) - host.cleanup() - rmSync(SCRATCH, { recursive: true, force: true }) - } -})