Repository navigation
OCR caution, scanner warning, table-fact labels and deleted-document titles (split from #3707) - #3715
Conversation
… deleted-document titles (split from #3707) Owner-approved split of #3707 (21 Oct 2026, 14:42 AWST). Moved unchanged from #3707 (Claude's commits af2dfc0, 2158725 caution part, 2abb635, 3752f2e, 4804078, 02bea13, 0f7b568): - scanned-page (OCR) source caution, kept inside the warning cap - worker start-up warning when the malware scanner module is not set, plus the decision record to keep the upload scan as is - table-fact merged row label carried into clinical_parameter (ING-B-09) - deleted document's title dropped from the storage clean-up ledger - the owner-approved temporary drift exception for source-governance tests
|
This pull request has been ignored for the connected project Preview Branches by Supabase. |
|
Warning Review limit reachedYour organization has reached its usage spending cap. Adjust your spending cap in the billing tab. Next included review available in 56 minutes. View limit details
Comment |
![gitar-bot[bot]](https://avatars.githubusercontent.com/in/827041?s=60&v=4){kind=small}
…r-tablefacts-titles
|
Owner approval (typed, relayed via Grok Bot, Oct 21 2026 15:46 AWST), verbatim: "I approve resealing #3715's hazard controls with the anchor refresh." |
…r-tablefacts-titles
…nalysis anchor refresh (source-governance.ts:364 -> :393) Owner approval typed 2026-10-21 15:46 AWST on #3715. Sealed with npm run governance:seal-hazard-controls.
|
Owner approvals (typed, relayed via Grok Bot, Oct 21 2026 15:49 AWST), verbatim:
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
BigSimmo has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Claude encountered an error after 2s —— View job I'll analyze this and get back to you. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4bc23745b4
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…r-tablefacts-titles
… merged cells; reset on empty rows
|
I approve restricting #3715's row-label carry-down to cells the parser marks as merged. |
…r-tablefacts-titles
|
Auto-merge disabled again: gitar-bot re-armed this after Helper 3 (the owner) disarmed it for the pending OCR-cap fix. Per the owner's rule, no bot re-arms a disarmed PR. Helper 3 will re-arm it once everything is in and green. |
|
I approve resealing #3715's hazard controls to include the grouped OCR warning that the warning cap never drops, and updating reviewedCommit/reviewedAt to that reseal. |
… drops (sealed path; needs owner-approved reseal)
…d reseal (comment 6313455205)
|
Auto-fix isn't included in your plan ↗ |
CI failed: Lighthouse budget check failed on three total-blocking-time regressions (desktop root, mobile document search, mobile root); the mobile breaches were confirmed 3/3 times, but the desktop-root confirmation was lost to a NO_NAVSTART runtime error. The analyzer judged this unrelated to the PR, which touches OCR, table-fact, and governance code, not the tested pages.OverviewOne unique error template across 7 analyzed logs, with a single failing job (119188672916). The failure is a Lighthouse performance budget breach on the FailuresLighthouse TBT budget regression on root and document-search pages (confidence: high)
Summary
Code Review ✅ Approved 1 closed / 1 findings🔴 High risk · Merged-cell carry-forward changes clinical table-fact labels for reindexed documents; incorrect span provenance could associate thresholds with the wrong parameter. Adds a grouped OCR caution that reserves a slot within the 8-warning cap, a worker start-up warning when the malware scanner module is unset, merged table-row label carry-down for newly indexed documents, and removal of deleted document titles from the storage clean-up ledger. No open findings; the change is approved. ✅ 1 closedClosed: Security: Reseal clears all 16 drift exceptions, not just the 3 approved files
VerificationRules No rules evaluated Functional validation Not enabledAuto-approval Approved For more verification details, comment Tip Comment OptionsAuto-apply is off → Gitar will not commit updates to this branch. Comment with these commands to change the behavior for this request:
Was this helpful? React with 👍 / 👎 | Gitar |
Conflicts were only EXTRACTOR_VERSION labels (PR raster-math-layout-v13 vs main merged-cell-provenance-v11); kept the PR's v13 label, which already postdates main's, with no further bump. Brings in #3715.
Split out of #3707 at the owner's request (21 Oct 2026, 14:42 AWST). This is part (b). #3707 keeps part (a), the Retry route and heartbeat.
Moved unchanged from #3707 (Claude's commits):
src/lib/source-governance.ts,src/lib/types.ts,tests/source-governance.test.ts)worker/admin-source-runtime.ts,worker/main.ts, runbook). Includes the decision record to keep the upload malware scan as is.clinical_parameter(worker/table-facts.ts).src/lib/ingestion.ts,scripts/cleanup-storage.ts, privacy impact assessment). This is the app-side half of the Clear deleted documents' titles from older clean-up ledger rows (database) #3706 backfill.tests/source-governance.test.ts(docs/clinical-hazard-controls.json).RAG impact: behaviour change. The OCR caution is governance only. ING-B-09 changes table facts for newly indexed documents only. The canary pair belongs around the re-index that first applies it.
Blocked: owner reseal needed (clinical)
Two unit tests fail, and an agent can't clear either:
clinical-doc-line-anchors:docs/clinical-hazard-analysis.md:93citessource-governance.ts:364forhasDangerSourceGovernanceWarning, which is now at line 393. Both files are sealed hazard-control paths, so refreshing the anchor changes a sealed digest.challenger-m5-2-empirical-stress: it expects at least 15 unexcepted sealed control paths, and the drift exception brings it to 14. Lowering that number would loosen a ratchet, which is not allowed.Clearing these needs the owner to re-review and reseal (
governance:seal-hazard-controls)src/lib/source-governance.ts,tests/source-governance.test.tsanddocs/clinical-hazard-analysis.md. The anchor refresh can then land with the reseal. The owner confirmed the wording on #3707 (05:18 UTC, 21 Oct). Stays a draft until then.Evidence (local):
table-facts,ingestion,worker-admin-source-runtime,source-governanceandworker-shadow-extractiontests pass. The only failures are the two above.Summary by cubic
Adds an OCR caution (not a refusal) when a cited source has pages read by OCR, a worker start-up warning when the malware scanner module is unset, a merged table-row label carried into
clinical_parameterfor newly indexed documents, and drops a deleted document's title from the storage clean-up ledger. The grouped OCR caution holds one reserved slot so the 8-warning cap can never drop it or crowd out document-specific warnings, and the decision record keeps the upload malware scan unchanged.Table-fact label carry-forward now inherits only into cells the Python extractor confirmed as covered by a merged span (PyMuPDF
Noneprovenance), and resets at an empty separator row, a row with its own label, a section heading, or a page/segment break. The owner-approved hazard-controls reseal landed with the anchor refresh and regenerated hazard-register snapshot, clearing the previously failingclinical-doc-line-anchorsandchallenger-m5-2-empirical-stresschecks.Written for commit 6d61d29. Summary will update on new commits.