starred/orca
1
0
Fork 0
mirror of https://github.com/stablyai/orca.git synced 2026-10-04 00:02:21 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(rpc): state the measured blindness, not the assumed one

Browse Source 
Applying each mutation to real product source shows the two holes are not equal.
The reorder is invisible to 83 of 84 tests and only a probe sees it. The refusal
blanking is also caught by the family reply matrix, because a refusal from cold
publishes null over a non-null initial value — an observational gap, not a
detection gap. Says so rather than letting the stronger claim carry both.

Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb
This commit is contained in:
Jinwoo-H
2026-09-13 18:53:26 -04:00
parent d932dab753
commit 7576360b2b
1 changed files with 11 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
mobile/src/test-support/rpc-recording/README.md
+11 -5
View File
@@ -149,11 +149,17 @@ rewritten call site produce the same sender calls, settlements, state and effect
It is not a substitute for reading the diff. Two facts bound it, both learned the hard way:
- **It was blind to refusal ordering.** Reordering the settings and sibling refusal checks in
`mobile-new-tab-agent-loader.ts` survived all 64 goldens as they shipped. A human reviewer caught
that class by reading #20499. Every pre-probe "refused" scenario refuses on the _first_ request,
and every correlated-failure schedule rejects at the transport, where neither check is reached.
- **It could not see data loss on refresh.** Same cause: with no prior success there is nothing for
a refusal to discard, so "does this screen keep its data or blank it?" had no observation.
`mobile-new-tab-agent-loader.ts` survives every golden except `probe-new-tab-both-refused` —
measured by applying the reorder to the real source: 1 failure in 84 tests, and the one failure
is a probe. Every pre-probe "refused" scenario refuses on the _first_ request, and every
correlated-failure schedule rejects at the transport, where neither check is reached. A human
reviewer caught that class by reading #20499.
- **It did not observe data loss on refresh.** No pre-probe golden records state after a refused
_refresh_, so "does this screen keep its data or blank it?" was undocumented. This one is an
observational gap, not a proven detection gap: publishing an unaccepted read in
`use-new-workspace-runtime-context.ts` is caught by the refuse-after-data probe _and_ by
`matrix-settings.workspace-context`, because a refusal from cold publishes `null` over a non-null
initial value. Claim the recorded behaviour, not blindness.
`probe-hole-witness.test.ts` closes both and keeps them closed. It asserts the hole and the closure
together: each probe must kill its mutation _and_ every pre-probe scenario of the same operation