test(ui-react): await accepted queue rendering in continuity checks

[schlessera]

Refs #1855

Hosted qualification blocked — 2026-10-18

Blocked by #1737
Blocked by #2386
Blocked by #2402

Ready run 39084453976, attempt 1, is complete: 13 executed jobs succeed; browser (2) and the aggregate fail. Packaging is intentionally unselected. All 15 executed checkout logs identify d2cb2d2a2f5e31e4a8a344dd23e16cd7faf79812, parents base 68e12f27eb249c9fe7262a498a0397c04bbda8f4 / head fc086b3498ed55a5423a41cab4e31eddd6dccd05, tree 136b600f9c19ae18c48961218cd577ac04db3abf; immutable Git API and local Git agree. All thirteen artifact archives are downloaded and API SHA-256 verified.

Browser (2), job 117548165246, has three independently attributed failures among 5,078 cases / 171 files:

• WebKit literal to grouped full notes capture light: unchanged 15,000-ms case deadline exceeded; report 15,429 ms. New ci: full grouped diagram capture exceeds WebKit case deadline #2402 owns this split full-width capture boundary, distinct from ci: authored narrow grouped diagram notes exceed WebKit case deadline #2386's authored-narrow correction.
• WebKit authored narrow grouped notes light: same 15,000-ms deadline exceeded; report 16,743 ms. Recurrence is recorded in ci: authored narrow grouped diagram notes exceed WebKit case deadline #2386; its independently owned test(ui-react): separate narrow callout geometry and capture #2393 remains untouched.
• Chromium rail-mixed, All commands by pointer and keyboard opens the palette, focus returns: dark, 480: original space opens assertion at desktop-navigation.pointer.tsx:329 sees null. Prior focus assertion passes; report 1,489.9 ms. This new natural occurrence is folded into ui-react: diagnose All commands Space activation after Escape in Chromium #1737 with no claim of native activation causation.

Both failing source files and desktop-palette production source match the tested base byte-for-byte. This four-file queue slice changes none of them. That source separation is not a causal diagnosis or permission to ignore the failures. The downloaded case report and raw log agree. Browser artifact 12120218335 has SHA-256 f9108327a3996cb12a015a5b5a4900ed3acac1dbd86bbad3a0f5515d82299149. The aggregate correctly reports browser: selected proof did not pass (failure); live main still equals the tested base. Browser (2) took 879 whole-job seconds; the whole failed confirmation cost 8,348 summed executed job-seconds. No unchanged retry or whole-CI saving is claimed.

The accepted-queue file itself passes all 15 native cases / 1,876 assertions, no skips, in 411.96 seconds, with exactly 87 observation calls / 87 retained frame calls. The current-head local preflight, original source-bound mutation controls and historical 9.703-second matched queue saving retain their own identities. Passing native evidence does not make the failed confirmation green.

#2188 is preserved as draft at fc086b3498ed55a5423a41cab4e31eddd6dccd05, release 0.44.0, with its original four-file Refs scope. Claim e1849-1855-r3-20261018 is released for this blocked handoff. No unrelated product/test fix, shortened observation, timeout increase, skipped assertion, unchanged shard retry or merge was performed. Original remaining wait-inventory and justified-horizon/equivalent-completion criteria stay open.

Resume after the named verification prerequisites have delivered qualified corrections or source-specific causal resolution: inspect live base and existing ownership; integrate relevant changes into this preserved slice; run required affected preflight and fresh complete selected confirmation; verify actual checkouts/artifacts/aggregate and release milestones before matching-head merge, then verify actual-main proof. No failed result is relabeled passing and no existing owner's branch is taken over.

Historical confirmation request — 2026-10-18

Current head fc086b3498ed55a5423a41cab4e31eddd6dccd05 integrates inspected main 68e12f27eb249c9fe7262a498a0397c04bbda8f4. Required guarded bun run check:pr --base origin/main passes all contribution gates, 54 fast tests / 181 assertions, and all 15 native continuity tests / 1,876 assertions, no skips, 525.41 seconds file elapsed. Exactly 87 observation calls and 87 retained frame calls remain, including both ten-cycle reconnect and IME cases. Diagnostic command elapsed is 571.05 seconds; this is qualification cost, not a performance comparison.

The first refreshed preflight stopped with exit 137 after 11 native cases passed, without a failed assertion. Kernel/user logs contained no corresponding OOM record and the task cgroup reported zero OOM events/kills. Its cause remains unproven. One bounded diagnostic execution added per-second owned-process identity/RSS and task memory/event sampling around the unchanged command. It completed successfully without an OOM event increase. The failed termination and successful diagnostic receipt are separate; no causal repair or permission to retry a real assertion failure is claimed. The preliminary diagnostic launcher could not find its optional timing binary and was corrected before launching tests. Original native deadlines, all guards, fixture logic and assertions remain unchanged.

Before requesting confirmation, independently queried main, PR head/base and fetched proposed merge agree: d2cb2d2a2f5e31e4a8a344dd23e16cd7faf79812, parents 68e12f27eb249c9fe7262a498a0397c04bbda8f4 / fc086b3498ed55a5423a41cab4e31eddd6dccd05, tree 136b600f9c19ae18c48961218cd577ac04db3abf. This requested candidate identity is not an executed hosted receipt. All selected current jobs, actual checkouts/artifacts and aggregate remain mandatory before milestone readback and matching-head merge. The earlier run 39082194681 remains failed for its correct stale-base refusal despite all fourteen other executed jobs succeeding.

The four-file accepted-queue Refs scope, source-attributed mutation controls and original remaining #1855 criteria remain unchanged. Release 0.44.0 and claim e1849-1855-r3-20261018 remain assigned during fresh confirmation.

Candidate refresh after stale-base refusal — 2026-10-18

Ready run 39082194681, attempt 1, has 14 successful executed jobs and a failed final aggregate. Packaging is intentionally unselected. All 15 executed checkout logs agree on d8c679f8da8a60726f9fa4bb41b7af90c28d400d, parents base 775eb17aab740fa759c2182f306c38fc17acbcbe and head 9c76ec888519c0c7788f18693d4f7324e7e61bee, tree f319072a9699705bd661f746a122303621bc7943; local Git and immutable API parents/tree agree. All thirteen artifact archives were downloaded and their API SHA-256 digests verified.

The final gate correctly reports Candidate became stale: the live PR head, readiness or base changed; request confirmation again. Independently queried main advanced to 68e12f27eb249c9fe7262a498a0397c04bbda8f4 during the run. Its sole new commit #2389 changes the worker runtime placement test to put its host sentinel behind the launcher's literal private /tmp mount even when TMPDIR is relocated. It changes no production behavior, continuity source, browser/layout assertion or guard. It still requires a new candidate confirmation; no stale-base bypass is justified.

All 15 native continuity cases / 1,876 assertions pass in hosted unit (1), 481.62 seconds file elapsed, with exactly 87 observation calls and 87 retained frame calls. The supplemental diagram qualification self-test passes in hosted unit (4), 12.861 seconds, without changing its 30-second bound. That is separate successful hosted evidence, not a diagnosis or repair of the earlier local deadline failure. Both browser jobs pass in 629/667 whole-job seconds, covering all 309 project/file pairs, 9,050 passes and the same ten existing pointer skips, with no unhandled errors. Both layout reports pass (504/701 cases). The failed confirmation cost 8,288 summed executed job-seconds; no whole-CI performance gain is inferred.

#2188 is draft during refresh. Head fc086b3498ed55a5423a41cab4e31eddd6dccd05 integrates the inspected main commit; its four-file accepted-queue scope remains unchanged. The initial refreshed preflight was interrupted; the bounded diagnostic preflight then passed as recorded above. Preserve the complete failed-run identity above. Under the current retention policy a failed overall source run cannot authorize retained browser/layout proof, so no passing-result cache or waiver is introduced. All selected fresh proof and final candidate identity must pass before matching-head merge. Release 0.44.0 and claim e1849-1855-r3-20261018 remain; #1855's original remaining criteria stay open under Refs.

What changed

The two ten-cycle continuity cases await native receipt minima and finite rendering for already-accepted queue reports, replacing twenty 500 ms observation windows. All 15 cases, 85 original assertion sites (1,836 executed expectations), both viewport paths and ten cycles per viewport remain. Forty scope/receipt expectations are added. The other 87 observation windows remain unchanged.

Why this shape

The fixture restricts this replacement to the synchronous pending-queue branch: no new draft acceptance, transcript adoption, turn transition or history refresh. Bounded per-payload receipts allow legitimate identical reports and prove application opportunities. The existing lower pending-stack test proves synchronous store application; real Chromium retains rendering, identity, focus, draft, upload and no-send assertions. Production code is unchanged. The guarantee distribution and source-bound study are in test-waits.md and test-wait-continuity-queue.json.

Proof

Prior local preflight passes at head 9c76ec888519c0c7788f18693d4f7324e7e61bee, integrating base 775eb17aab740fa759c2182f306c38fc17acbcbe. The four-file PR diff remains unchanged. The queue handler, connection/draft paths and both native test files remain byte-identical to the prior qualified head b088ec26703a270e8295c0f70aa82e096d576a11. Main's maintenance identity, native phase-deadline and diagram-assertion changes have been inspected; they do not replace current-candidate proof. The measured browser-weight prerequisite #2366 is delivered by #2379 with separate complete ready and actual-main proof.

A supplemental current-source tests/native-diagram-inputs.test.ts run completed all 13 assertion calls but failed its existing 30-second test deadline (30.378 seconds). No failed expectation was printed. Its source and installed qualification fingerprint are unchanged; this is distinct from #2365's source-mismatch failure. Cause is unproven, no unchanged retry or deadline increase was made, and fresh hosted unit proof must pass. This failed local receipt is retained separately from required preflight.

Prior required local preflight: all contribution gates, 54 fast tests / 181 assertions, and all 15 native continuity tests / 1,876 assertions pass (521.36 seconds file elapsed), with exactly 87 observations and 87 frame calls retained. This qualifies the current source locally; hosted proof is still pending.

The immutable historical controls in the study retain their original source/runtime attribution:

• The same production unstable-id fault fails the unchanged removed-pill assertion with both original and replacement waits.
• A two-second native delivery delay is awaited; bypassing receipt completion fails on missing payloads before delivery, including with the final duplicate-tolerant predicate.
• An extra identical report in every phone cycle passes all ten cycles.
• Deferring the production report write through queueMicrotask fails the existing immediate pending-stack assertion; restoring it passes the full lower file (20 cases / 86 expectations).

The historical matched pair on frozen base 7d1a31d09b96739a85bff5151cc2e1300c9ac4d0 preserves 15 cases / 1,836 original expectations, adding forty checks. Twenty queue sections measure 10.973 to 1.270 seconds: 9.703 seconds observed saving. Command time is 419.73 to 402.45 seconds, including setup/shared-host variation; CPU is 123.55 to 119.88 seconds and peak RSS 1,247,680 to 1,244,888 KiB. This is not a whole-CI speedup. Eleven study commands including deliberate faults cost 1,608.39 driver-wall / 1,119.76 CPU seconds. Strict-equality measurement and final minimum-predicate qualification remain distinguished in the source-bound evidence.

Historical ready run 39070164840 remains failed: browser shard 1 exceeded its 15-minute whole-job bound during finalization, so the aggregate failed despite passing test output. Its checkout f86c5b4005eb4954a51f391752189130d20bfbb9 has parents 0d780605fe1ea9a281cce947b1f0074f3f3e7acc and b088ec26703a270e8295c0f70aa82e096d576a11, tree 9351a4654654bb56ef933c11e3f97a0c64ca45e4. All 15 executed checkout logs agree. That failed run cost 8,411 summed job-seconds / 1,042 event seconds. Earlier failed qualification, corrections and source-attributed measurements remain in the issue history; no failed or older receipt is current green evidence.

Checklist

• Focused local preflight and complete selected hosted proof passed, with actual head/base/checkout identities recorded.
• Completed slice and remaining original acceptance criteria are named below.
• PR and advanced issue retain confirmed upcoming release 0.44.0 (milestone 58); re-read before merge.
• Tests/docs only; no consumer-visible changeset or contract change.
• Leakage and citation gates apply to the whole diff.
• Documentation describes the changed guarantee distribution.
• Provider-facing history is unchanged.

Anything left open

This is the accepted-queue slice only. #1855 stays open for caller-specific justified horizons or fault-sensitive equivalent completion of the remaining 87 windows and original inventory criteria. Their retained durations are not established minima. No production timing policy, offline guard or negative assertion is weakened. Complete selected hosted confirmation, actual checkout/aggregate reconciliation, milestone readback and matching-head merge remain required for this candidate. Proposed merge d8c679f8da8a60726f9fa4bb41b7af90c28d400d has the recorded base/head parents and tree f319072a9699705bd661f746a122303621bc7943; actual selected jobs must independently verify their checkout.