Skip to content

Draft: integrate all 225 frozen open Transcripted PRs - #2530

Merged
r3dbars merged 487 commits into
mainfrom
codex/integration-open-prs-20261110
Nov 12, 2026
Merged

r3dbars merged 487 commits into
mainfrom
codex/integration-open-prs-20261110

Conversation

@r3dbars

@r3dbars r3dbars commented Nov 10, 2026 •

Copy link
Copy Markdown
Owner

Combined candidate

Combines 225 frozen PR heads into one draft integration branch for combined-app testing. All exact heads and original contributor history are retained; main, original branches and releases are unchanged. The original 198 and additional 27 are frozen separately so later arrivals do not silently expand the candidate.

The integration ledger and manifests record every PR, dependency, disposition and conflict resolution. Decision worksheets remain proposals; README privacy wording is preserved separately rather than adopted.

Integration fixes reconcile overlapping async Writing work, audio lifecycle ownership, library migration and list behavior. Independent reviews found an obsolete scroll-smoke initializer, missing full-gate CI checks, historical aliases hiding distinct co-speakers, and an exception-path leak in the new bounded diagnostic reporter. Those findings are fixed and recorded with their regression evidence. The current narrow fixture follow-up was checked by the parent; no formal GitHub approval or exhaustive native acceptance is claimed.

Validation

Current head: 58b89c633fdf002d014c7b35b8f4149e85bb4e92. Production app sources are unchanged from owner-tested 5840b0e.

  • Latest fixture corrections: 22 focused Core methods PASS; audio graph component reproduced 2 failures before correction and passes 282/282 assertions after. Mutation controls still catch missing watchdog cancellation, missing old cleanup and successor shutdown. These test-only changes restore the inherited logger-disable flag and use real successor fake resources; production audio sources are unchanged.
  • Previous head 872154b: quick suite 112 PASS / 0 FAIL / 0 SKIP; Core test targets compile. Its full hosted bench remains FAIL in fast/Core suites. App build, E2E, synthetic pasteback, integration, imported audio, QA round-trip/stress and synthetic audio passed; artifact validation had one warning.
  • The replacement full bench and ordinary Swift CI must qualify the current exact head. Prior-head passes, syntax parsing, focused components and skipped jobs do not substitute for that result.
  • Bounded failure reports publish only tracked repository Swift identifiers and line numbers; raw test/assertion content and absolute paths remain private. Six reporter tests pass, including the Python 3.9 symlink-loop regression.

Main promotion remains held

Needs successful exact-head CI/full bench, isolated native UI/import verification and physical mic/system-audio/AirPods/pasteback checks. Stats corruption-recovery behavior and licensing discrepancies still require resolution. No signed release, updater publication or production health clearance is claimed.

Exact-head hosted and owner-run proof

Full bench 43007027969 PASS at 5840b0e4233f62aac334949de389a2b002181b16: exit0, all15 expected steps recorded,14PASS/1non-blocking artifact-validationWARN. Actual app build, fast tests, Core tests, E2E, synthetic pasteback, integration, imported audio, QA tests/round-trip/stress, synthetic audio and fixture release-health checks all pass. Earlier failed runs remain failed historical evidence. This is developer-app qualification, not signed-release clearance.

Owner manual testing of the same source build on this Mac: dictation paste reported working; a meeting with both channels logged2mic/1system utterances and saved; subsequent route-change test logged3mic/3system utterances with zero mic gaps/dropped callbacks and a saved transcript. That route-change meeting used the built-in mic, so AirPods-microphone loss/recovery is NOT proven. Owner reports repeated Load More, timestamp playback, Tab autocomplete, stale-suggestion dismissal on cursor movement, and quit/reopen with test meetings still listed/playable all PASS. These are owner reports plus bounded local log evidence, not automated native acceptance.

Ordinary PR Swift CI is restarted as attempt2 of run43006972744 on the same 5840b0e source. Repository MAC_RUNNER_MODE is now off, selecting GitHub-hosted Macs; the stranded Mac-runner attempt was force-cancelled before rerun. Package/native scroll checks and app-build outcomes remain pending. Existing Stats-recovery acceptance, licensing and remaining physical/native coverage holds remain. Main and public releases are unchanged.

Capture-list fixture follow-up

Ordinary hosted CI attempt2 of43006972744 reached the native capture-list smoke and failed during fixture admission, before testing the lists. The guard equated Foundation's system temporary directory with explicitly supplied TMPDIR; a standalone Foundation probe reproduced the mismatch. Head58b89c6 uses the explicit canonical fixture directory plus UID/type/0700 checks, rejects symlinks/traversal/unsafe leaves, and preserves the native-account preflight, volatile capture override and exact-leaf deletion guards. All36 filesystem promises pass, with mutation controls for the old temp equality and omitted permission check. Parent independently reviewed the full patch and reran its regression.

This is test infrastructure only: the running owner app and production Sources are unchanged. Clean exact-head quick suite at58b89c6: 113PASS/0FAIL/0SKIP, exit0 and RESULT PASS, including the36 actual Swift fixture-admission promises. Ordinary hosted CI and the full bench remain pending on this head. Historical5840 fullbenchPASS remains valid for that earlier head, not a substitute for replacement-head native-scroll proof.

Owner decision: Stats corruption preservation accepted

Owner explicitly accepted preserving a corrupt Stats database with Stats unavailable until manual repair/reset. Recordings and saved transcripts remain available; there is no repair UI or automatic backfill. The #2491 product-decision hold is resolved for this integration.

Read-only technical provenance review of #2387 found no factual correction needed in the MCP and Parakeet notices: they retain the upstream transition/discrepancy rather than selecting a broader grant. Model-artifact provenance and upstream licensing resolution remain separate distribution requirements; this is not release/legal clearance. Other decision worksheets remain unselected proposals.

Main promotion still waits for current-head CI/native-scroll proof and the remaining explicitly identified verification requirements.

Native scroll callback correction

The spm-tests job on 58b89c633fdf002d014c7b35b8f4149e85bb4e92 passed fixture admission, then aborted in the scroll smoke with Swift's runtime exclusivity violation. Its diagnostic closure read the same step variable held by an active inout parameter. The correction passes that diagnostic value explicitly; list interaction, assertions, fixture isolation and production code are unchanged.

A standalone Swift reproduction fails with the original captured read (SIGABRT / simultaneous accesses) and passes with the value parameter, retaining the expected diagnostic sequence. The edited smoke syntax-parses. Independent static review is clean and confirmed the other callback captures do not alias the borrowed values. This is not native UI execution; replacement hosted proof remains required. The correction is committed as fdb0236ae0da9df0204c80d044e290fd489b9690.

The full bench for the preceding 58b89c6 source passed: run 43019791968, trusted workflow 8a38235eef00816a79daa1324387a21ef880204d. Downloaded source receipt matches the complete SHA: 14 PASS, 1 nonblocking synthetic artifact validation WARN, all 15 expected steps recorded, exit 0. It does not supersede the separate failed native scroll smoke. Promotion remains held for replacement-head hosted validation and remaining real-hardware evidence.

Callback-fix validation: bash check.sh quick passed with 113 checks, zero failures/skips. verify-change.sh --base HEAD reported no applicable app check for the test-only diff. Worktree is clean and the remote matches fdb0236ae0da9df0204c80d044e290fd489b9690. Replacement hosted full bench accepted as run 43055348415 for that exact source SHA, hardware false, staged source empty; last observed queued. Normal push CI and native scroll results remain pending.

Exact-source native scroll proof — PASS

Corrected genuine Actions run 43076762597 validated source fdb0236ae0da9df0204c80d044e290fd489b9690 in a fresh isolated macOS 26 VM, using diagnostic workflow 658b8aa94d8334f1845c1e8ec87355ad9d13237d. Original native isolation guard, dependency build, native compilation, 36 fixture-path assertions, completion receipt and all four scenarios passed. Dictations/Meetings at 80 and 150 captures reached the bottom; no growth, scroll-jump, layout-churn or hang assertion failed. Receipt/artifacts: stage=complete, exit=0.

The first diagnostic run 43076339916 remains a failed setup attempt: an extra ownership-chain check used for the protected Library marker was incorrectly applied to the macOS runtime stamp directory. Correcting only that diagnostic check preserved the original native isolation guard and all candidate code.

Owned VM, ephemeral runner and private share were removed; original VMs remain stopped. Owner app/data and MAC_RUNNER_MODE=off unchanged. This is focused native proof, not full-suite, full-bench or hardware acceptance. Original hosted normal CI and full bench remain queued at last readback.

Full-check findings and locale correction

The local isolated Actions validation of preceding source fdb0236 completed 29/30 full-check stages successfully: 404/404 fast-test functions and 54,244 assertions; Core XCTest reported 2,048 tests, zero failures, 17 skips, plus 1,136 Swift Testing tests. All four Lab commands passed. Strict concurrency census reported six warnings (Speech 4, Meeting 2); regression versus compiler differences is unresolved pending same-toolchain comparison with base 8a38235.

The literal full bench on that source remained FAIL: 10 PASS, 2 FAIL, 3 WARN, all 15 expected stages recorded. Dependency validation produced different hashes for identical inputs under C and en_US.UTF-8 locales, causing build/integration failures and blocking imported-audio proof. The run also had a shallow-checkout preflight warning. Artifact upload/finalization failed; complete receipts and raw logs were recovered locally rather than claimed as published evidence.

Replacement source e66e3e4fb7c4e73e5a6825a8f66458425ca12d16 fixes only deterministic dependency path sorting and adds a cross-locale regression. The regression failed before the fix and passes after it; both locales now return the same digest. Validation: 19 dependency freshness contracts, 43 toolchain contracts, archive-input contracts and independent static review PASS. bash check.sh quick: 113 PASS, zero failures/skips. verify-change.sh --base HEAD: no applicable app check for this build-tooling/test change.

Fresh exact-source full bench and head/base concurrency comparison are in progress through the isolated CI route. No baseline increase, warning suppression, production-source modification, main merge or release. Owner app/data remain untouched.

Actions service startup blocker on e66e3e4

Checks are configured: Swift CI 43097548657, hygiene 43097548741, census 43097548653, and Lab 43097548873 were all created for exact source e66e3e4fb7c4e73e5a6825a8f66458425ca12d16, but remain queued without reported check results. The exact-source hosted full bench 43097997107 returned startup_failure before creating any job. Isolated diagnostic attempts 43097734750 and 43098161623 also failed at Actions startup with zero jobs and explicit unexpected-service-error annotations. These establish no test pass or failure.

No further blind dispatch retries are being made. The independent local VM compiler-only head/base comparison and locale verification are ongoing; they do not substitute for a completed native full bench. The existing local 113-check PASS and earlier exact-fdb native scroll/full-check evidence remain scoped to their stated sources. Promotion remains held.

Xcode 26.5 head/base comparison — PASS, no warning regression

Candidate e66e3e4fb7c4e73e5a6825a8f66458425ca12d16 and actual base 8a38235eef00816a79daa1324387a21ef880204d independently rebuilt their own dependencies and typechecked in one fresh local VM using identical Xcode 26.5 / Swift 6.3.2 toolchains. Immutable Git archives were verified inside the guest and source bytes remained unchanged after compilation.

Both revisions produce the same six concurrency diagnostics at identical source locations: MeetingAudioFileServices.swift:259,274; AppleSpeechEngine.swift:643,647; DictationPreviewSampleSink.swift:87,91. Meeting=2 and Speech=4 for both. Strict zero-baseline check remains FAIL; the dedicated workflow's same-toolchain --allow-up-to base comparison PASSES. No baseline or source warning was changed or suppressed.

The locale fix also passed all 19 freshness contracts on Xcode 26.5; C and en_US.UTF-8 produce identical dependency digests. Both dependency rebuilds and report-mode censuses passed. All owned VMs/shares were cleaned up and original VMs remained stopped. This compiler-only proof used no fabricated Actions identity and ran no native app.

Local source/command/diagnostic/cleanup evidence: /tmp/transcripted-compiler-e66-20261111-run/REPORT.md. Exact replacement-head native full-bench proof remains pending due Actions service startup failures; no merge/release acceptance is claimed.

November 12 validation retry

At the owner's request, restarted validation without source or runner-routing changes. Exact candidate remains e66e3e4fb7c4e73e5a6825a8f66458425ca12d16.

  • Normal Swift CI 43197629401: actual candidate head verified, workflow started; first Linux routing job in progress at readback. Mac test/build results pending.
  • Full bench 43197633425: dispatched from trusted main with exact candidate source input, hardware false and staged source empty; accepted and pending. Its workflow SHA is the trusted main SHA, not the tested candidate SHA; final source receipt remains required.

Cancellation of stale run 43097548657 was refused by GitHub as already completed even though its read API showed queued. No unrelated run was canceled. Previous failure and proof records remain intact; no new test success claimed yet.

November 12 exact-head results

Both validations finished on the exact candidate e66e3e4fb7c4e73e5a6825a8f66458425ca12d16.

  • Normal Swift CI 43197629401: PASS. checks, spm-tests, app-build and build-and-test all succeeded.
  • Full bench 43197633425, attempt 2: PASS. The receipt records source_sha e66e3e4f…, bench_exit_code 0, and one non-blocking artifact-validation WARN, the same one as on 5840b0e. Attempt 1 was cancelled by GitHub after it waited six hours for a hosted Mac and never got one; it ran no steps.

Independent review of the candidate against main (16 area reviewers, each finding checked by a second reader): core audio, meetings, speech, UI, MCP/CLI and the cross-PR merge check came back clean. One P1 in release tooling (the release-candidate diff guard doesn't allow the new BuildVersion.swift that the bump script writes, so the next release prep would fail) and about 14 minor items (license-notice gaps, test-matrix rules, a stderr handler, a library-move retry case) are being fixed on claude/intfix-* draft PRs against this branch.

Still not proven: AirPods microphone loss and recovery on real hardware, and native UI checks on a separate account.

…istently (TOOLS-CLI-6)

# Conflicts:
#	Tools/TranscriptedCLI/Sources/TranscriptedCLI/ContextStore.swift
#	Tools/TranscriptedCLI/Tests/TranscriptedCLITests/ContextStoreTests.swift
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:36 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:37 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:41 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:41 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:42 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:42 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:43 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:43 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:44 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:44 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:44 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:45 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:45 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:46 — with GitHub Actions Active
@r3dbars
r3dbars deployed to automerge-ai November 12, 2026 22:46 — with GitHub Actions Active

This branch was successfully deployed

1 active and 1 inactive deployments
automerge-app — edbc761f Deployed Nov 13, 2026 by r3dbars via gate #3259
automerge-ai — edbc761f Deployed Nov 13, 2026 by r3dbars via review #3259
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant