docs: record phase 6 re-audit - #100
Merged
Merged
Conversation
Uh oh!
There was an error while loading. Please reload this page.
BigSimmo pushed a commit
that referenced
this pull request
Jul 30, 2026
Addresses the Codex P2 on this PR and fixes my own mistake in the first attempt.
The correction itself is unchanged in substance: `documents_status_id_idx` is
canary-gated, not ordering-safe, because `rag-candidate-sources.ts:482` pairs
`.eq("status","indexed")` with an unordered `.limit(12)` and `(status, id)`
serves that equality. The reasoning stays in the runbook, which is the
authority; the ledger row now just points at it.
What changed is the shape. My first version added prose to #102's Detail cell,
which is the widest cell in that column — so Prettier repadded every row and the
diff became 63 insertions / 63 deletions. Row #133 records that this exact
whole-table behaviour has already caused mechanical conflicts that silently
stopped all CI, and it is very likely how the original correction was lost
between #1423 and main in the first place.
Three attempts established the rule empirically, since my first two hypotheses
were both wrong: growing the cell reflows (68 lines), *shrinking* it also
reflows (68 lines, because the column width tracks this cell), and trailing
whitespace padding does not help because Prettier trims it. The replacement is
therefore exactly the same length as the text it replaces — 97 characters — and
the diff is now 1 insertion / 1 deletion with Prettier reporting the file
unchanged.
Verified on an untouched row: before this, #100's Detail cell lost exactly one
padding space (3533 -> 3532 chars), which is what proved the column width, not
my cell's content, was driving the churn.
Gates: outstanding-issues guard 137 rows / unique ids; branch-review-ledger 117
live + 1206 archived, no duplicates; prettier clean. No SQL applied.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01361jh3eYVjJCzXWjAhdZiFBigSimmo added a commit
that referenced
this pull request
Jul 30, 2026
…#1440) * docs(issues): re-apply #102's canary-gated correction lost to a merge Codex raised this as a P2 on PR #1423: with the runbook corrected, ledger #102 still told an operator the `(status,id)` use was ordering-safe, so the two operator entry points contradicted each other. A reply on that thread claimed the fix landed in 2bdc245. It did not reach `main`. Verified by content rather than by SHA, which matters here: #1423 was squash-merged, so `git merge-base --is-ancestor` returns NO for every original commit and proves nothing either way. Reading `docs/outstanding-issues.md` at `origin/main` (0310c6a) shows the retracted sentence still present, while the runbook correction and the test change from the same PR both landed. Most likely cause, and it is a documented hazard: #1428 ("archive five completed ledger rows") touched this same file on `main`, and #112 records that `docs/outstanding-issues.md` conflicts present as ordinary content conflicts a hurried resolution can settle by dropping one side. Union merge has since been added to this file, but it would not have saved this case — both sides edited the same sentence. The row now mirrors the runbook's classification and names the runbook as the authority if the two ever diverge again, so a future reader has a tie-breaker rather than two equal-looking claims. Gates: verify:cheap exit 0, Test Files 434 passed (434), Tests 4563 passed | 4 skipped. Outstanding-issues guard, branch-review-ledger guard, docs:check-links, docs:check-scripts, prettier all pass. No SQL applied; no index created. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01361jh3eYVjJCzXWjAhdZiF * docs(issues): correct #102 within the existing table width Addresses the Codex P2 on this PR and fixes my own mistake in the first attempt. The correction itself is unchanged in substance: `documents_status_id_idx` is canary-gated, not ordering-safe, because `rag-candidate-sources.ts:482` pairs `.eq("status","indexed")` with an unordered `.limit(12)` and `(status, id)` serves that equality. The reasoning stays in the runbook, which is the authority; the ledger row now just points at it. What changed is the shape. My first version added prose to #102's Detail cell, which is the widest cell in that column — so Prettier repadded every row and the diff became 63 insertions / 63 deletions. Row #133 records that this exact whole-table behaviour has already caused mechanical conflicts that silently stopped all CI, and it is very likely how the original correction was lost between #1423 and main in the first place. Three attempts established the rule empirically, since my first two hypotheses were both wrong: growing the cell reflows (68 lines), *shrinking* it also reflows (68 lines, because the column width tracks this cell), and trailing whitespace padding does not help because Prettier trims it. The replacement is therefore exactly the same length as the text it replaces — 97 characters — and the diff is now 1 insertion / 1 deletion with Prettier reporting the file unchanged. Verified on an untouched row: before this, #100's Detail cell lost exactly one padding space (3533 -> 3532 chars), which is what proved the column width, not my cell's content, was driving the churn. Gates: outstanding-issues guard 137 rows / unique ids; branch-review-ledger 117 live + 1206 archived, no duplicates; prettier clean. No SQL applied. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01361jh3eYVjJCzXWjAhdZiF * docs: record issue correction review --------- Co-authored-by: Claude <noreply@anthropic.com>
This was referenced Jul 30, 2026
BigSimmo pushed a commit
that referenced
this pull request
Aug 13, 2026
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WBdo125Dh3idPcF7CmCCmG
BigSimmo pushed a commit
that referenced
this pull request
Aug 13, 2026
… preview Incremental verified delivery per docs/verified-answer-incremental-delivery-design.md: - answer-stream-contract: VerifiedUnit types and isDeliverableVerifiedUnit (schema version, kind allowlist, strictly-increasing sequence, bounded JSON size); the token/revising SSE event names remain excluded. - answer-preview: buildEvidencePreviewUnit reuses the production danger-level source-governance refusal and the exact route-boundary source trim; any danger warning suppresses the preview entirely (stricter than the final response, whose refusal only fires on grounded supported answers). - answer-progress-public: optional verifiedUnit passthrough, dropped (never repaired) when validation fails; old clients ignore the field. - rag.ts: the retrieved progress event carries the preview only when RAG_INCREMENTAL_EVIDENCE_PREVIEW=true (new env flag, default false in env.ts and .env.example). No new SSE event name; Phase 2 per-section generation is not attempted (provider-gated by design). - tests/answer-incremental-delivery.test.ts: Phase 0 offline contract proof — schema/sequence validation, token/revising rejection, governance refusal emitting zero units, owner-boundary trimming, byte-identical preview/final reconciliation at the trim layer, flag-off default. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WBdo125Dh3idPcF7CmCCmG
BigSimmo pushed a commit
that referenced
this pull request
Aug 27, 2026
… panel Direction B from /mockups/answer-loading-redesign, applied to the live surface after the clinical owner chose it (2026-08-27). `AnswerProgressStepper` is gone. It was a filled accent panel carrying a 36px icon tile, a five-circle stepper with connecting rails, a scrolling ECG trace, a per-second elapsed counter and a Processing details disclosure — roughly 210px of the loudest chrome in the product, occupying the answer's own position for the four to twelve seconds before every answer, on a surface that PRs #2386 and #2388 had just spent their whole argument quietening. It narrated the orchestrator's five stages, which the reader is not operating, and it never showed a single source. `AnswerProgress` replaces it with one status line plus the arriving source rail. The rail is the substantive half. The evidence preview already crossed the stream boundary before the prose — trimmed, owner-scoped, governed, consumed by the client — and was being rendered as a SECOND full panel below the stepper, with its own icon tile, heading and three-column card grid. Two loud blocks in the answer's position, both removed when the answer arrived. It is now a horizontal rail of small cards drawn to look like the source rail the arrived answer renders, mounted inside AnswerProgress rather than beside it, so nothing is removed and nothing jumps when the answer lands. Cards carry a dot, not a number. The preview is the top slice of retrieval in retrieval order; the final list is rebuilt from what the answer cites and re-capped by trust. A number assigned during the wait can point at a different document once the answer lands, which is the precise failure the citation design exists to prevent. Pinned by a DOM test, and stated in the rail's accessible name for a reader who never sees the dot. Checked while building, not assumed: `isDeliverableVerifiedUnit` pins evidence_preview to sequence 0 and rejects a repeat, so the preview crosses once, complete — it cannot accrue card by card. And it is gated behind NEXT_PUBLIC_RAG_INCREMENTAL_EVIDENCE_PREVIEW_RENDER (#100 Phase 1), off by default, so today the line carries the accrual alone via resultCount and the rail is simply absent rather than a placeholder. Improvements beyond the mockup: - Retrieval counts passages and selection counts sources, never one noun for both. Collapsing them is how a reader believes two dozen documents are behind an answer that cites three. Pinned in tests/answer-progress.test.ts. - `fallback` says "Assembling the answer from the sources directly" while it is happening. Twenty of thirty answers in the 2026-08-18 blinded read were source_only; the wait is the honest place to set that expectation. - Processing details became "How this answer was built", shown only after a retrying/fallback/cached run. The old disclosure held the same five stages for every question, which is why nobody opened it. - The elapsed counter is gone. In its place one threshold at 10s appends "taking longer than usual" and does not tick. Nothing can be done with the number while the search is healthy; re-drawing it every second in the position the eye rests on is what made the wait the subject. - SearchProgressBanner (library/document modes) loses its filled accent band and spinning Loader2 for the same line. Fill is how this app marks a hazard. - AnswerSkeleton drops its own status line. Found in the browser, not in a test: it renders in the answer's body slot directly under AnswerProgress, so the screen showed "Writing the answer…" above "Reading your question…". There is now exactly one place that says what is happening. - The completion dot is not green. A status hue that nothing else on the element repeats is a colour-only signal, and it was redundant beside a line already reading "Answer ready in 3s". Caught by check:design-system-contract. Motion. The ECG trace, its two animation tokens, its keyframes and its compositor rules are deleted; the indicator is a 5px dot breathing on opacity. That choice is what holds the contract ui-phone-motion.spec.ts exists for, after Reduce Motion set the trace to opacity 0 and left a dead panel on a physical iPhone: a stopped dot is a complete, correct bullet, where a stopped spinner is a fragment of a circle. The animation stays in globals.css rather than a motion-safe: utility because html[data-motion="full"] must be able to opt back in over the OS, which a Tailwind media variant cannot express. tests/answer-activity-trace-css.test.ts pinned the deleted component, so it is replaced by tests/answer-progress-indicator-css.test.ts carrying the same regression forward against the dot — including an assertion that the trace stays deleted, since a partial revival is how the original defect shipped. docs/search-chrome-behaviour.md's physical-iPhone rubric named the ECG strip by class and would have sent a tester hunting for markup that no longer exists; docs/design-system/COMPONENTS.md named the component in its live-region adoption note. Both updated — the second found by check:dead-code-candidate, which is the gate doing its job. Verified: verify:cheap exit 0 — 37 gates enforced, design-system contract passed (1021 production files, no ratchet moved), 886 test files / 10733 tests passed; Playwright chromium answer-progress-ui-smoke + ui-phone-motion 9 passed, ui-universal-search 20 passed; driven in a real browser at 390px through scoping → retrieving → retrieved(24) → ranking(4 AU, 2 WA) → generating and past the 10s threshold, and again with the preview flag on to confirm six real cards render, link to /documents/<id>?page=&chunk=, and carry no numbers. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012kHT2YKCNnUrVckTaJW6ga
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Verification
npm run test -- tests/private-access-routes.test.ts tests/source-governance.test.ts tests/source-text-sanitizer.test.ts- 77 passednpm run test:e2e:chromium -- tests/ui-smoke.spec.ts -g "search regressions avoid fetch errors and open viewer hits"- 1 passednpm run verify:cheap- runtime, lint, typecheck, 583 Vitest tests passednpm run verify:ui- 35 Chromium Playwright tests passednpm run verify:releasebefore release or handoff confidence claimsnpm run format:checknpm run check:production-readiness- passed with expected local env warnings onlynpm run check:supabase-project- passed against Clinical KB Database (sjrfecxgysukkwxsowpy)npm run check:indexing- passed; 2065/2065 indexed, 0 failed jobs, 0 missing embeddings, 0 mixed-generation documentsnpm run reindex:health- passednpm run reindex:cleanup-staged- dry run found 0 eligible documents and 0 artifact rowsClinical Governance Preflight
Clinical KB Database(sjrfecxgysukkwxsowpy)Notes
scripts/classify-documents.tswas left unstaged and is not part of this PR.npm run check:indexingneeded a longer local timeout when run alone; it passed in 369s.