mirror of
https://github.com/stablyai/orca.git
synced 2026-09-30 16:02:56 +00:00
test: register late submission settlement reliability gate
This commit is contained in:
@@ -10,6 +10,84 @@
|
||||
}
|
||||
},
|
||||
"gates": [
|
||||
{
|
||||
"id": "agent-session.submission-settlement",
|
||||
"title": "Late Claude delivery proof settles the original unknown submission",
|
||||
"maturity": "experimental",
|
||||
"protection": "partial",
|
||||
"owner": "native-chat",
|
||||
"layer": "runtime-integration-and-mutation-contract",
|
||||
"surfaces": ["Claude structured chat", "submission journal", "shared structured outbox"],
|
||||
"platforms": ["macos", "linux", "windows", "ios", "android"],
|
||||
"providers": ["local", "remote-runtime", "relay"],
|
||||
"coveredPlatforms": ["macos"],
|
||||
"coveredProviders": ["local"],
|
||||
"coverageNotes": "Production runtime, host, journal, subscription and shared outbox with scripted Claude transport and process identity probe on macOS; folder workspace exercised. Other native hosts and paired clients are relevant but not covered by this run. Direct SSH/WSL structured Claude execution remains unsupported by existing location checks.",
|
||||
"motivatingLinks": ["https://github.com/stablyai/orca/pull/19141"],
|
||||
"invariant": "A correlated late user echo settles only its original unknown submission accepted, clears its unconfirmed outbox state, suppresses stale retry delivery and preserves a newer active turn.",
|
||||
"oracle": "Timeout the first send, accept a newer send, deliver the first user echo, then require accepted journal/subscription state, removed outbox entry, two provider sends despite explicit retry, and successful newer-turn cancellation. Mutation contracts require serialization behind unknown persistence and one write/publication across repeated settlement.",
|
||||
"commands": [
|
||||
"ORCA_BACKGROUND_LAUNCH=1 pnpm test src/main/native-chat/agent-session-wire/structured-agent-session-late-dispatch.test.ts src/main/runtime/claude-late-dispatch-integration.test.ts"
|
||||
],
|
||||
"testFiles": [
|
||||
"src/main/native-chat/agent-session-wire/structured-agent-session-late-dispatch.test.ts",
|
||||
"src/main/runtime/claude-late-dispatch-integration.test.ts"
|
||||
],
|
||||
"assertionRefs": [
|
||||
{
|
||||
"file": "src/main/native-chat/agent-session-wire/structured-agent-session-late-dispatch.test.ts",
|
||||
"assertions": [
|
||||
"waits for the send to persist unknown before accepting and publishing once",
|
||||
"ignores delivery proof after the host session has closed"
|
||||
]
|
||||
},
|
||||
{
|
||||
"file": "src/main/runtime/claude-late-dispatch-integration.test.ts",
|
||||
"assertions": [
|
||||
"settles a timed-out unknown through the runtime, publishes outbox removal and fences retries"
|
||||
]
|
||||
}
|
||||
],
|
||||
"evidenceRuns": [
|
||||
{
|
||||
"date": "2026-09-06",
|
||||
"runner": "local",
|
||||
"platform": "macos",
|
||||
"command": "ORCA_BACKGROUND_LAUNCH=1 pnpm test src/main/native-chat/agent-session-wire/structured-agent-session-late-dispatch.test.ts src/main/runtime/claude-late-dispatch-integration.test.ts",
|
||||
"result": "passed",
|
||||
"durationSeconds": 15.45,
|
||||
"summary": "7/7 passed after correcting the scripted PID lease-probe fixture; product source at 88d478071a. Earlier broad runs had a process-inspection failure and the fixture failure; no wholly green final broad run claimed."
|
||||
}
|
||||
],
|
||||
"runtimeBudget": {
|
||||
"p95Seconds": 30,
|
||||
"scope": "Target for this focused two-file command; observed run 15.45 seconds, measured p95 not established."
|
||||
},
|
||||
"flakeHistory": {
|
||||
"status": "unknown",
|
||||
"evidence": "One corrected focused 7/7 run; sustained soak history missing. Earlier fixture lease-probe failure was corrected, not retried away."
|
||||
},
|
||||
"redGreenEvidence": {
|
||||
"status": "complete",
|
||||
"evidence": "Removing onLateDispatchAccepted runtime adapter forwarding fails the integrated journal assertion with unknown/accepted instead of accepted/accepted; restoring forwarding passes. PR 19141 review artifacts retain ablation and restored-run logs."
|
||||
},
|
||||
"performanceBudget": {
|
||||
"required": true,
|
||||
"evidence": "Mutation contract counts one write/publication across repeated settlement; integration counts two provider sends despite retry. No added polling/timers/processes; exceptional settlement copies/scans session submissions. Large-journal latency and subscriber fanout are unmeasured."
|
||||
},
|
||||
"promotionCriteria": [
|
||||
"Collect the policy-required soak runs and days without unexplained flakes.",
|
||||
"Retain forwarding ablation sensitivity, send/write/publication counts and newer-turn cancellation assertions.",
|
||||
"Establish native host and paired-client coverage before claiming those platforms protected."
|
||||
],
|
||||
"knownGaps": [
|
||||
"Scripted provider integration does not itself prove rendered Electron behavior or live Claude ordering.",
|
||||
"No native Windows/Linux, native iOS/Android, pinned mixed-version client, or paired transport reconnect run for this oracle.",
|
||||
"No settlement-write failure/restart fault injection or release-scale journal performance measurement.",
|
||||
"Existing oldest-unconfirmed Retry targeting is outside this fix."
|
||||
],
|
||||
"demotionRule": "Remain experimental on unexplained failures or lost ablation sensitivity; do not mask failures with retries, skips or longer timeouts."
|
||||
},
|
||||
{
|
||||
"id": "ssh.localhost-terminal-agent-hooks",
|
||||
"title": "Localhost SSH terminal and agent hooks reach the owning pane",
|
||||
|
||||
Reference in New Issue
Block a user