Skip to content

Integrate OCR 1.11.5 with private tool diagnostics - #178

Merged
xeonvs merged 9 commits into
mainfrom
codex/v0.9.1-ocr-1.11.5
Sep 7, 2026
Merged

Integrate OCR 1.11.5 with private tool diagnostics#178
xeonvs merged 9 commits into
mainfrom
codex/v0.9.1-ocr-1.11.5

Conversation

@xeonvs

@xeonvs xeonvs commented Sep 6, 2026

Copy link
Copy Markdown
Owner

Result

Integrate OCR 1.11.4 and 1.11.5 for toolkit 0.9.1, with exact OCR 1.11.5 as the supported runtime. Bounded private failed-tool arguments are discarded before normalized diagnostics and publication DLP, while valid review signal and GitLab summaries remain publishable. Tracks #176 and #177 in milestone v0.9.1.

Changes

  • Accept bounded opaque failure_details.arguments without exposing it in finalized results, receipts, notes, telemetry, or public DLP decisions.
  • Preserve valid findings and summaries when optional diagnostics are malformed; keep public-field DLP and later-action decisions independent.
  • Fail OCR qualification closed on malformed JSON, non-object result/comment/tool-call/detail shapes, or non-string selected preview paths through controlled CompatibilityError / contracts/contract-probe-failed status.
  • Qualify OCR 1.11.4 comment repair and Objective-C++ routing separately from OCR 1.11.5 grouping/dependency/viewer changes; retain adjacent evidence and target only 1.11.5.
  • Keep live qualification forward-only, with isolated frozen historical readback and version-neutral current operator guidance.
  • Route AGENTS.md and toolkit strategy to the canonical Mermaid decision flow covering diagnostics, DLP, receipts, publication, errors, warnings, success, and later actions.

Exact head

  • Commit: b47fa551b5de42dcfe9cd5b316a32f4184525be6
  • Tree: f934ed54598f1ff4696ef5f93c0e7ceb9f279f70
  • Runtime subtree: 754b35fd0de50ed684e78ccfce9012dec6f0b888
  • OCR Linux amd64 SHA-256: 53a4ab7c8ce6dc07d5362c7c4984bf8d98b55e4e8d4c01b9399d488a2a983d95
  • OCR Darwin arm64 SHA-256: c041b03cc840957b52df28514e8dbb51f798e6cb1259d97555a41a2e3e3ccaf9
  • Upstream checksum-file SHA-256: 0519c13b03d69dd6c4aa5470a8eb52727f5e3fd2ce43e5887d8a67cdf107aa1a

Validation

  • Final local gate: 1537 tests + 408 subtests, 86.40% combined coverage; risk groups 85/82/86/88%. Ruff, runtime MyPy, Bandit, lock, manifest/evidence validation, Towncrier, diff checks, and pinned Gitleaks 8.24.3 passed.
  • Checksum-verified OCR 1.11.4 and 1.11.5 passed deterministic no-LLM compatibility probes; hosted qualification evidence remains linked from the issues and manifest.
  • Configured model-backed OCR 1.11.5 completed immutable range ae0a9ac..029cdf0: selected/completed 9/9, failed/reused/waived 0/0/0, 77 tool calls, no GitLab posting. The two malformed-output findings were fixed and regression-tested; retained private evidence SHA-256 is 3a5374a846c3033734da0c66d7a9032e54043e96b8a8322fd9724f37691c2873.
  • Codex Security scan 5db2da0c-3f84-46a4-b4dd-cf3521fd1402 closed 7/7 security-relevant surfaces through 57f5236; incremental scan 90628f3c-10aa-4ce9-a3fa-b8c497e504d4 covered 57f5236..ecd7ecb. Both completed with no reportable findings and emitted sealed reports/SARIF.
  • Full commit-chain self-review and exact-head pre-push review passed. OCR diagnostic artifacts are retained owner-only for audit; no provider configuration, credential, raw output, or GitLab mutation is published.

Release path

This PR is the completed feature gate. After exact-head hosted checks and thread review, make it Ready and squash-merge. Then verify the protected-main TestPyPI development publication, prepare the separate protected Release v0.9.1 PR, complete stable TestPyPI/PyPI/tag/provenance/attestation/immutable-Release readback, close #176/#177 and the milestone, clean branches, synchronize main, and finish the no-release external reconciliation without producing another stable release.

@xeonvs

xeonvs commented Sep 6, 2026

Copy link
Copy Markdown
Owner Author

Draft handoff receipt

Implementation and local/hosted verification are complete in Draft PR #178.

  • Commit: d570195a595a8cf33bdde2d7a7d909e7d7e4eabe
  • Tree: da6e85558919db80ddad5ed7ec2faded24a969b1
  • Runtime subtree: 4fb9322348850340cf97a5b5ea99d03166ef6f5c
  • All 13 reported PR checks passed, including both macOS jobs, Linux Python 3.12–3.14, packages, dependencies, Security and CodeQL.
  • Final OCR qualification: https://github.com/xeonvs/open-code-review-toolkit/actions/runs/34023700224 — passed. Downloaded Linux contracts and asset hashes exactly match the committed expanded 1.11.5 evidence.
  • Local: 1524 tests + 408 subtests; 86.39% coverage; all four risk groups passed. Real no-LLM OCR 1.11.4/1.11.5 contracts were identical before/after qualification refactoring.
  • Verified private-arguments/DLP/publication equivalence through the real toolkit finalization and posting workflow with captured GitLab transport.
  • Qualification maintenance instructions are now canonical and routed from AGENTS.md; live probes no longer select behavior by historical OCR versions, while archived evidence readback remains isolated.

The final workflow is a regression run against the already-promoted Draft manifest. Original adjacent 1.11.3→1.11.4 and 1.11.4→1.11.5 audit identities remain in the two separate committed evidence files and manifest conclusions.

This completes WQ-06's local and hosted delivery portion. Real LLM/model validation remains pending in another configured environment. Follow the Draft body checklist using this PR's built toolkit wheel and exact OCR 1.11.5, not the published 0.9.0 wheel. PR remains Draft and mergeable, with no unresolved threads; issues #176/#177 and milestone v0.9.1 remain open. No merge or package publication was performed.

@xeonvs

xeonvs commented Sep 6, 2026

Copy link
Copy Markdown
Owner Author

Version-neutral documentation follow-up completed at signed commit 029cdf09cc01d47035c0cf37772c922d9027d2ce (tree f2f8180a05700ea4301ae384cba7b01cec39212c). Current operator guidance and generic fixtures no longer duplicate OCR release numbers/checksums; manifest references preserve exact installation requirements. Executable pins, current-pin assertions, runtime and historical evidence are unchanged. Canonical maintenance instructions and documentation regression tests cover the separation.

Self-review completed before commit. Local quality: 1525 tests + 408 subtests, 86.39% coverage and all risk-group floors; Ruff, MyPy, Bandit, lock, manifest, Towncrier, diff and pinned Gitleaks checks passed. GitHub independently verifies the commit signature. All 13 checks on this exact head are SUCCESS, including all five OS/Python jobs, package build, dependencies, Security and CodeQL. CI run: https://github.com/xeonvs/open-code-review-toolkit/actions/runs/34025051012 . Local and remote heads agree; worktree is clean.

WQ-07 and the WQ-06 local/push/hosted portion are complete. The PR remains Draft and mergeable; no merge, release or issue closure was performed. External model qualification remains pending. Use the Build artifacts wheel for this head. The runtime subtree remains 4fb9322348850340cf97a5b5ea99d03166ef6f5c; prior real no-LLM OCR qualification remains attributable to its recorded predecessor, not a newly executed model test.

@xeonvs
xeonvs marked this pull request as ready for review September 7, 2026 08:01
@xeonvs

xeonvs commented Sep 7, 2026

Copy link
Copy Markdown
Owner Author

Final feature gate

  • Exact reviewed head: b47fa551b5de42dcfe9cd5b316a32f4184525be6
  • Tree: f934ed54598f1ff4696ef5f93c0e7ceb9f279f70
  • Runtime subtree: 754b35fd0de50ed684e78ccfce9012dec6f0b888
  • Hosted checks: 13/13 completed successfully
  • Unresolved review threads: 0
  • Local completed-head gate: 1537 tests and 408 subtests; 86.40% combined coverage; risk groups 85/82/86/88%; Ruff, MyPy, Bandit, lock, compatibility evidence, Towncrier, diff checks and Gitleaks passed
  • Model-backed OCR 1.11.5: immutable range ae0a9ac..029cdf0, selected/completed 9/9, failed/reused/waived 0/0/0, no GitLab posting; retained owner-only result SHA-256 3a5374a846c3033734da0c66d7a9032e54043e96b8a8322fd9724f37691c2873
  • Codex Security: scans 5db2da0c-3f84-46a4-b4dd-cf3521fd1402 and 90628f3c-10aa-4ce9-a3fa-b8c497e504d4 closed 8/8 combined review surfaces with zero reportable findings

The feature branch is ready for protected squash merge. Stable publication and external reconciliation remain separate release lifecycle states.

@xeonvs
xeonvs merged commit de9055e into main Sep 7, 2026
13 checks passed
@xeonvs
xeonvs deleted the codex/v0.9.1-ocr-1.11.5 branch September 7, 2026 08:04
@xeonvs xeonvs mentioned this pull request Sep 7, 2026
xeonvs added a commit that referenced this pull request Sep 7, 2026
## Scope

Prepare and authorize stable toolkit **v0.9.1** for the completed OCR
1.11.4/1.11.5 qualification and diagnostic-safety work tracked by:

- #176
- #177

This release PR contains repository-side release preparation only.
Stable publication and external reconciliation remain pending until this
exact PR is merged and the protected Release workflow completes.

## Reviewed implementation

- Feature PR: #178
- Reviewed feature head: `b47fa551b5de42dcfe9cd5b316a32f4184525be6`
- Protected squash merge: `de9055e699ff8a7bc0806c6eae5bdf6666e95160`
- Release base: `de9055e699ff8a7bc0806c6eae5bdf6666e95160`
- Release head: `4a3847af59b7597975e0daeb1f2867cee3077d01`

## Development publication

- Workflow:
https://github.com/xeonvs/open-code-review-toolkit/actions/runs/34098626798
- Version: `0.9.1.dev87`
- Wheel SHA-256:
`2cb519f4e54b25bf1e64416fdacd17ccbba7eba99a758ec0437bbd3021ba2829`
- Sdist SHA-256:
`9872ea9fcb8c81fd42264af325705e717f311265becffeab192e6ec28c07fd0e`
- TestPyPI PEP 740 verification and wheel/sdist install smoke passed.

## Review receipts

- Local OCR 1.11.5 reviewed `ae0a9ac..029cdf0`: 9 selected, 9 completed,
0 failed/reused/waived, 77 tool calls, no GitLab posting; result SHA-256
`3a5374a846c3033734da0c66d7a9032e54043e96b8a8322fd9724f37691c2873`.
- Codex Security scans `5db2da0c-3f84-46a4-b4dd-cf3521fd1402` and
`90628f3c-10aa-4ce9-a3fa-b8c497e504d4`: 8/8 combined surfaces closed, 0
reportable findings.

## Release validation

- 1,538 tests plus 408 subtests passed.
- Combined coverage: 86.40%; locked risk groups: 85% / 82% / 86% / 88%.
- Ruff, MyPy, Bandit, compatibility validation, release-note extraction,
`uv lock --check`, diff checks, and pinned Gitleaks passed.
- Two source-epoch-controlled builds were byte-identical.
- Wheel SHA-256:
`c1343b2eba77378d2c4a143bad445e47a762e0e67e3a916cb8f69e2f748f45fb`
- Sdist SHA-256:
`32ef050255aa9dc0056935bb4280fde9ace0490df0a0affb82a98655f14ce994`
- Twine and clean package metadata/content checks passed.

## Release preparation

- Stable marker: `0.9.1`
- Next development marker: `0.9.2`
- Deterministic source epoch: `1788768251`
- Authorized issue set: `[176, 177]`
- Towncrier fragments rendered into `CHANGELOG.md`.
- Public GitLab example pinned to toolkit `0.9.1`.
- Complete execution plan archived at
`docs/engineering/execution_history/releases.md#plan-toolkit-0-9-1` with
stable external delivery pending.
- `.release-reconciled-version` intentionally remains `0.9.0` until
independent external readback is complete.

## Post-merge gates

The protected Release workflow must independently prove registry bytes,
PEP 740 provenance, GitHub attestations, annotated tag target, immutable
GitHub Release and exact assets, release receipt, supported-Python
registry installs, Actions-owned issue receipts, issue/milestone
closure, release-branch cleanup, and synchronized `main`. A separate
protected `no-release` reconciliation PR will then record those facts
and advance `.release-reconciled-version` without creating or mutating
another stable release.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[OCR compatibility] Qualify v1.11.4

1 participant