fix: discard retired capture and settle observations

[thymikee]

Summary

Capture attempts now read the latest record of their admitted session lifetime, including recovery attempts. Publishing a snapshot creates a new record, preserving previously captured record identity.

Optional settle observations are discarded after retirement, including point-touch responses that await a recording frame probe. Tap corroboration compares the latest pre-probe baseline. Cached snapshots cannot authorize a retired lifetime.

Thirteen files changed. Part of #3116, stacked on #3145; addresses capture and settle findings on #3141.

Validation

Validated 0b76607bc7: 85 focused tests, quick and Fallow checks. Eight planted regressions fail; a ninth pre-fix control reproduces stale point-touch settle refs (1 failed | 10 passed), then all 11 response tests pass. The strengthened diff control also fails when snapshot assignment is omitted; the old control passed that regression.

pnpm check:affected --base fix/open-session-lifetimes --run passes all selected runnable gates and 1,636 related tests across 265 files. Independent review found no remaining findings. CI and live-provider evidence remain pending on the published head.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.96 MB	4.96 MB	+279 B
Package (unpacked)	4.96 MB	4.96 MB	+279 B
Package (download)	1.49 MB	1.49 MB	+115 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	25.1 ms	25.1 ms	+0.0 ms
CLI --help	75.0 ms	75.5 ms	+0.5 ms

[cubic-dev-ai]

All reported issues were addressed across 13 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[thymikee]

Thanks for the fix. The retirement logic in 0b76607 looks right, but I need a live run of the two device routes it changes before this can merge.

snapshot-command-runtime.ts:149 now publishes every snapshot and diff on an existing session through sessionStore.update(ref, {}). Settle-carrying touches now issue refs through resolveCurrent instead of requireCurrent. Every real session takes both routes, and only unit fixtures cover them so far. The PR body says live-provider evidence is still pending, and the PR guidelines ask for a live run of the changed path. Please run one session on an iOS simulator or Android device at 0b76607. Run open <app>, then snapshot -i, then press @eN --settle, which must return settle.refsGeneration. Then press a settle-issued ref with press @eM, which must resolve with no stale-ref error. Then run record start, click <x> <y> --settle, which must return settle.refsGeneration, and record stop. Please paste the JSON responses. The unit tests already cover the retirement race, and a live run cannot reliably reproduce it.

Not blocking, and you can take or leave these: SelectorCaptureRuntimeParams.session in selector-capture-runtime.ts is now unread, yet find-target-capture.ts and selector-runtime-backend.ts still pass it, so it could go with both caller arguments. The snapshot transition at snapshot-command-runtime.ts:149 mutates the record after commit, so update(ref, (current) => ({...})) would be safer. interaction-touch-reference-frame.ts:75 re-resolves a device platform that cannot change within a lifetime. The diagnostic churn could also be reverted.

On the open threads, the cubic-dev-ai P2 on the non-point touch outcome still applies. No stale settle is published on either path. Still, press @ref and press x y end differently for the same retired state, so please answer which outcome you want.

Smoke Tests is still in progress. Its device route goes through the snapshot publication and settle ref issuance this diff changes, so a failure there needs attributing before anyone dismisses it. I did not rerun the tests or check retirement races in the swipe, scroll or fill routes. Separately, finalizeTouchInteraction still records the action on the retired session object. That is outside this diff and fits the later #3116 slice. Before merge, I need the live evidence above and an answer on the thread.

[thymikee]

Completed the live control at exact head 0b76607bc7 on the owned emulator-5580, after rebuilding the checkout and both Android helpers.

The fresh interactive snapshot reported android-helper, version 0.21.18, generation 547198. Pressing its Network & internet ref with --settle returned a settled partial frame at generation 547202, including Navigate up. The next request pressed @e2~s547202 directly and succeeded, returning home at generation 547205, without another snapshot. The bare @e2 negative control was correctly refused before dispatch with plain_ref_requires_complete_frame and the exact pinned-ref hint.

Then started touch-enabled recording, clicked the previously observed Settings row coordinates (540, 739) with --settle, and stopped recording. Key actual responses:

{
  "refPress": {"success": true, "ref": "e11", "refLabel": "Network & internet", "settled": true, "refsGeneration": 547202},
  "nextPress": {"success": true, "ref": "e2", "refLabel": "Navigate up", "settled": true, "refsGeneration": 547205},
  "recordStart": {"recording": "started", "recordingBackend": "adb screenrecord", "showTouches": true},
  "pointClick": {"success": true, "targetKind": "point", "x": 540, "y": 739, "settled": true, "refsGeneration": 547210},
  "recordStop": {"recording": "stopped", "recorder": "confirmed", "nativePathDisposition": "retired", "showTouches": true}
}

The exported video is playable H.264, 1080×2400, 84.58 seconds. The attached screenshot shows the Network & internet page reached by the recorded point action. The owned session and isolated daemon were cleaned afterward.

[thymikee]

Thanks for the live run at 0b76607. It covers both changed routes: press @eN --settle returned a settle generation, the settle-issued ref then resolved with no stale-ref error, and the recorded click x y --settle returned its generation before record stop confirmed the video. The open question on the non-point touch outcome is answered by #3151. The code verdict is clean and there are no conflicts, so this is ready for human review.

[thymikee]

Consolidated into #3144 as part of reducing #3116 to seven PRs. The composition preserves the complete pre-consolidation source tree, including tests and later review corrections. This PR is superseded; its review discussion and native evidence remain available. Outstanding findings transfer to the owning keeper in the implementation record.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-03 21:13 UTC