The automatic main workflow 39877819113, at unchanged main b18e1d63d344452cf45719cf3b4072a752b71bce, failed the check job in ./buddy test. This is the automatic post-publication run, not a rerun. The exact-head PR workflow and local full suite had passed; those successes do not replace this failure.
The Linux suite reported 1,540 passed, 41 skipped, 1 failed, 489,784 assertions, 1,582 tests across 200 files, 3,158.55 seconds. Firmware, workspace-browser, execution-browser and purge-namespace jobs all succeeded.
The failed case is native execution registration guard shares the sole mutation writer and deadlines in tests/unit/public-safety-execution-registration.test.ts. Its native child exited 1 after 16,402 ms. The inner error is Actual SQL COMMIT precedes the scoped deadline at tests/fixtures/public-safety-access/execution-registration.ts:216, inside the native afterCommit hook. The surrounding assert.rejects(..., status(404)) saw that assertion instead of the expected refusal. This is not the 90-second outer child watchdog.
The fixture iterates result, token and incident deadlines, but the retained failure does not identify which iteration failed or the timestamps at SQL COMMIT versus hook entry. A production deadline defect, setup expiry, delayed hook observation and host scheduling remain unproven. No policy relaxation or arbitrary timeout increase is justified.
The original terminal log was downloaded exactly once. Its cached original archive is retained: 328,187 bytes, SHA-256 f78298238d8bde50f2c604792a6503db58af9a1432af27ba0af0829e47f785f5. The complete 639-line failed group is also retained separately. No workflow rerun has been requested.
Acceptance:
- Retain exact source, failed original workflow, child error and complete original archive.
- Reproduce or characterize the original case at a bounded native test seam; preserve every outcome and runtime identity.
- Distinguish each deadline setup, actual SQL completion, hook entry and response release with bounded observations. Identify the failing scope and causal timing before changing behavior.
- Correct the established cause without weakening deadlines, rollback, authority, persisted-state or refresh-hint assertions.
- Qualify the correction at the original native case and current-source full CI; preserve the original failure separately.
Related historical full-suite investigations: #60 and #68. No causal equivalence is claimed.
Controlled native characterization and published correction
PR #75, head b46d9c13d4271af82d54033d4cac5f8bf7847abe, contains the fixture correction, now published to main in 37ad383afae41f17d40ec08329bc989de448dc3c with three separately reviewed documentation corrections. Its exact-head full CI passed all five jobs.
A real 600 ms pause in the installed native transaction callback reproduces the original fixture assertion after its 500 ms checked deadline. A separate native SQLite connection observes exactly one committed source and version 2. A diagnostic counterprobe at the same actual boundary observes the application's correct 404 and exactly one refresh hint. Original production bytes were restored after each probe; the original Linux scope and precise scheduling point remain unknown.
The correction replaces the assumed commit-before-expiry precondition with an independent committed-visibility oracle and explicit phase classification. It retains all result/token/incident lifetimes, scoped 404 expectations, precommit rollback/no-hint checks, postcommit source/audit/version/hint assertions and the outer watchdog. A fixed fourth case deliberately crosses expiry during the native callback; at least one ordinary case must also establish a genuine response held past expiry. These are fixed distinct cases, not retries until passing. Bounded scalar receipts now propagate through the native test harness on success and failure.
Sensitivity proof: temporarily removing the sole production post-COMMIT result check makes the corrected native case fail with Missing expected rejection. Exact production bytes were restored. The corrected original native case passes all four scope/phase cases; Buddy Pickier and application/runtime types pass. Independent Spec and Standards reviews are clean.
The exact-head Linux suite passed 1,541 tests, 41 skips, zero failures and 489,786 assertions across 200 files. Original-archive readback verifies all four registration receipts: result/token/incident cases demonstrate a response held after native return and released after expiry; the fixed crossing case reports expiry at native return. All four observe refusal and exactly one committed source. Four copied standalone UDP/TCP protocol campaigns also pass, without firmware or physical execution claims.
Published by normal fast-forward to main as Chris Breuer, without coauthor trailers; #75 is merged. The automatic post-publication main workflow is a separate run and remains in progress.
Keep this issue open for the remaining unchecked historical-scope/timing acceptance item. The original failed archive does not identify its iteration or exact native chronology. Controlled reproduction, the fixture correction and passing qualification do not recover those unavailable observations or establish physical vehicle acceptance.