docs(issues): reopen #343 and #318, correct stale RAG coordination state - #2206
Conversation
Applies #Q5JHBJ's Phase 6.2 completion update plus three new rows: OpenAI ZDR contradiction, docling-lab-fixtures.v2, and the live-drift PGRST106 blocker. The #343 re-add and #318 reopen requests queued earlier in this branch are left pending: check:ledger-write-discipline requires a reconciled request to already exist on origin/main, so those two ride a follow-up reconcile after this PR lands.
§7 still described a hypothetical "Wave 2: S2 once the post-#2088 canary is green" and "waiting on canary approvals for S1c, S1d, S2" — all three landed and their canaries went green on 2026-08-18/19. Rewrite to match HANDOVER.md's current status table: Track A complete through S3 (#2108), Gate B PASS (#2154), S7/B4 merged (#2170, shadow defaults to legacy), S8+ still owner-gated. State what the owner actually still owes: the Gate E blinded read, the B4 shadow-mode production enable decision, B5's Gate A sign-off, a B6 go/no-go, and the three pinned adversarial divergences (#C2D9JF, #NTAV3D, #VXB8XA).
This pull request has been ignored for the connected project Preview Branches by Supabase. |
Warning Review limit reachedYou’ve reached a temporary PR review limit under our Fair Usage Limits Policy. Next review available in:35 minutes Limit details: You’ve used the included review currently available. Your 86 included PR review attempts over the past 7 days set your current allowance at 1 review per hour. Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab. How can I continue?Wait for the limit to reset, then comment An organization admin can change what happens after included review limits in Billing. How do review limits work?CodeRabbit enforces per-developer PR review limits within each organization. For paid Pro and Pro+ reviews, CodeRabbit uses a developer's included PR review attempts over the past 7 days to set the current hourly allowance. At typical activity levels, the full plan allowance applies. Higher sustained activity can lower the allowance until earlier attempts leave the 7-day window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (12)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
CI triageCI failed on this PR. Automated classification of the 2 failed job(s):
Compared with main CI run #12662 (success). Classification is evidence routing, not permission to ignore a failure. Exact quarantined Playwright identities remain governed by the flake ledger. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit:4e9e31279d
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".
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Immutable review record for claude/ledger-reconciliation-docs-truth-b0a2e9 at 4e9e312.
…t landed while this PR was open)
A new P3 task landed on main after this PR's earlier reconcile: the B4 shadow-extraction Gate F runbook (docs/worker-deploy-runbook.md section 3) has four operator unknowns the repo can't answer -- Railway variable-change restart behaviour, no read path for documents.metadata.shadow_extraction, an unratified 10% timeout rollback threshold, and unverified worker memory headroom. This lands after the main-sync merge that brought it in; the #343/#318 requests queued earlier in this branch are still deliberately left pending for the same reason recorded in the earlier reconcile commit.
…nt base The reconciliation on this branch was computed against an older base. `main` has since advanced (3ed1932) and its own edits to docs/outstanding-issues.md mean the branch's ledger is no longer byte-identical to a clean application of the queued requests from the current base, which is what check:ledger-write-discipline audits — it failed with "does not exactly match the audited application of 5 moved inbox request(s) from the base". This commit restores the canonical ledger and all five base inbox requests to their state at the current base so the reconciliation can be re-run as one clean transaction. No request content changes; the two requests this branch adds (39b3c784, b9a9492c) are untouched and stay pending for a later run. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…liation-docs-truth-b0a2e9
…rent base Re-runs the reconciliation as a single transaction from base e92a1b1, so docs/outstanding-issues.md is byte-identical to the audited application of the five moved inbox requests. That is what check:ledger-write-discipline verifies, and it is what the previous stale-base run could not satisfy once main advanced. Applied: 069da585, 219680d3, a20fc4ce, bbac85a9, cc60253d. The 219680d3 request is the docling shadow-extraction runbook follow-up queued by PR #2200, which was the request the earlier partial batch left behind and the reason CI was red. The two requests this branch raises (39b3c784, b9a9492c) stay pending by design: they are not part of the base inbox, so a later reconciliation owns them. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Both P2 findings on PR #2206 were verified against the ledger and are correct. #VZN8G3 duplicates the earlier, still-open #BSBE9B: both require the same docling-lab-fixtures.v2 corpus (unruled tables, merged/spanning cells, rotated headers) and both gate a docling table-quality promotion argument on it. The duplicate came from inbox request a20fc4ce being queued without checking the ledger for an existing row, and was only visible once reconciliation allocated an id. #BSBE9B stays canonical. #Q5JHBJ was left in Open items while its own detail records Phase 6.2 as complete with proof, so completed deployment and guard work could be re-dispatched. Its summary's deliverables are all evidenced in the row, and the row itself states the single residual is out of its scope and tracked as #M54C4N. These are queued as immutable done requests rather than edits to the canonical ledger, so they apply at the next reconciliation. They cannot be reconciled in this PR: the write-discipline audit compares the ledger against a clean application of the requests present in the PR's base on main, and these two were created on this branch. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Uh oh!
There was an error while loading. Please reload this page.
…ger (#2218) Applies the 29 immutable inbox requests pending at base af2075a as one audited transaction, so docs/outstanding-issues.md equals a clean application of that batch and check:ledger-write-discipline can verify it. Includes the two closures queued during the PR #2206 review, which could not be reconciled there because they were created on that branch rather than present in its base: - #VZN8G3 closed as a duplicate of the earlier, still-open #BSBE9B (same docling-lab-fixtures.v2 table-hardness corpus, same promotion gate). - #Q5JHBJ archived: its own detail recorded Phase 6.2 complete with proof while the row stayed in Open items, leaving finished deployment and guard work re-dispatchable. Its single residual is tracked separately as #M54C4N. The remaining 27 requests were queued by other sessions for exactly this step and are applied verbatim; the reconciler validates each against the row it targets and rejects stale ones. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Summary
source_metadataretrieval-row pin was made "structural," but PR feat(ledger): record 15 delivered resolutions, harden operator runbooks, and relax source metadata schema #2107 loosened it, PR fix(rag): restore the strict source_metadata presence pin on retrieval rows #2121 restored the strict Zod-level pin, and the database-sidecheck (jsonb_typeof(metadata) = 'object')constraint the closure referenced was never added (confirmed viagit logand a grep ofsupabase/schema.sql/supabase/migrations/). The new row (#S19JRT) carries the actionable follow-up and the verified evidence.#1YPV51): its 2026-08-18 closure claimed the Clinical Lead had completed the medication-interaction-lexicon review and sign-off. The repo owner confirmed in chat that this did not happen — the lexicon remains clinically unreviewed.npm run issues:reconcilefrom a fresh base, applying the two rows above plus three already-pending requests (OpenAI ZDR-status contradiction,docling-lab-fixtures.v2, and the live-drift PGRST106 blocker) and a completion update to#Q5JHBJ.docs/rag-improvement/COORDINATION.md, which still described "Wave 2: S2 once the post-revert(rag): back out #2065 claim-leading condition binding after live canary regression #2088 canary is green" as pending — that work landed weeks ago. Rewritten to matchHANDOVER.md's current status table (Track A complete through S3/feat(answer): compose follow-up chips from the S2 menu, gate them on evidence (packet S3 / A4) #2108, Gate B PASS/Packet S6b: Gate B PASS — Docling lab benchmark run, decision record, and harness fixes #2154, S7/B4 merged/feat(worker): packet B4 — docling shadow extraction mode (WORKER_DOCUMENT_EXTRACTOR_MODE=legacy|shadow, default legacy) #2170) and to state what the owner actually still owes (Gate E blinded read, B4 shadow-mode production enable decision, B5's Gate A sign-off, a B6 go/no-go, and the three pinned adversarial divergences).Clinical Governance Preflight
Not applicable — this PR only touches
docs/outstanding-issues.md, its inbox request/audit files, anddocs/rag-improvement/COORDINATION.md. No ingestion, answer-generation, source-governance, privacy, production-env, or clinical-output code changed.RAG impact
RAG impact: no retrieval behaviour change — this PR edits only the RAG programme's status-tracking coordination doc (
docs/rag-improvement/COORDINATION.md§7) to match already-landed, already-merged state recorded inHANDOVER.md. No code undersrc/lib/rag/**or any protected retrieval/ranking surface changed.Test plan
npm run check:branch-review-ledger— passednpm run check:ledger-write-discipline— "Ledger write discipline passed for a1c2ced..HEAD."npm run verify:pr-local— all 11 selected checks completed, 0 failed, 0 not reached (docs-only scope; lint/typecheck/unit suite/build correctly skipped as low-risk)npm run format— clean across the whole repo, 0 files needed a fix🤖 Generated with Claude Code