Merge branch cleanup and review workflow updates

[BigSimmo]

Summary:

• Integrates the LLM pipeline review cleanup branch, PR430 UI test stabilization, and Codex review throttling protocol into main.
• Refreshes the Supabase drift manifest after merging schema-adjacent changes.
• Keeps the merge path through PR because direct pushes to main are blocked by repository rules.

Verification run:

• npm run verify:cheap: passed once after drift manifest refresh; after the final remote-main merge, one full rerun timed out in tests/public-access-deep.test.ts, and that same file passed on focused rerun.
• npm run test -- tests/rag-routing.test.ts tests/registry-corpus.test.ts tests/clinical-search.test.ts tests/retrieval-query-variants.test.ts tests/rag-answer-fallback.test.ts: passed, 153 tests.
• npm run test -- tests/drift-detection.test.ts: passed, 10 tests.
• npm run test -- tests/public-access-deep.test.ts: passed, 5 tests.
• node scripts/run-playwright.mjs tests/ui-smoke.spec.ts --project=chromium -g "document viewer": passed, 4 tests.
• node scripts/run-playwright.mjs tests/ui-tools.spec.ts --project=chromium: passed, 40 tests.

Checks not fully green:

• npm run check:production-readiness is blocked locally by missing NEXT_PUBLIC_SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY, and OPENAI_API_KEY.
• Full npm run verify:ui failed once late in the suite; the failing document-viewer area passed on targeted rerun.

Risks:

• Main is active and moved during this cleanup, so this PR should rely on required branch checks before merge.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e2b23df6e2

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".

[BigSimmo]

@copilot resolve the merge conflicts in this pull request

[coderabbitai]

Actionable comments posted: 4

🧹 Nitpick comments (4)

src/lib/registry-corpus.ts (1)

431-448: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Minor duplication with skippedRegistryEmbedResult.

Line 446 reconstructs the "skipped" result shape inline instead of reusing the helper defined at Line 356.

-    return { documentCount: 0, chunkCount: 0, skipped: true, reason: "failed", errorMessage } satisfies RegistryCorpusEmbedResult;
+    return { ...skippedRegistryEmbedResult("failed"), errorMessage };

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/registry-corpus.ts` around lines 431 - 448, The fallback result in
bestEffortReembedRegistryRecordAfterEdit duplicates the skipped
RegistryCorpusEmbedResult shape instead of reusing the shared helper. Update the
catch path to construct the failure/skip response through
skippedRegistryEmbedResult (the helper used in this module) so the result shape
stays centralized and consistent, while still preserving the errorMessage from
embedError and the existing console.error context in
bestEffortReembedRegistryRecordAfterEdit.

tests/registry-corpus.test.ts (1)

74-98: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Test only covers the "medication" dispatch branch.

reembedClinicalRegistryRecordBySlug/reembedDifferentialRecordBySlug each independently re-check registryCorpusEmbeddingEnabled(). Consider parameterizing this test over "service"/"differential" targets too, so a future dispatch branch that forgets the disabled-check isn't missed.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/registry-corpus.test.ts` around lines 74 - 98, The disabled-embedding
test in registry-corpus only exercises the reembedRegistryRecordAfterEdit
medication path, so it can miss regressions in other dispatch branches. Extend
the existing test around reembedRegistryRecordAfterEdit to parameterize over the
service and differential flows as well, covering
reembedClinicalRegistryRecordBySlug and reembedDifferentialRecordBySlug so each
branch is verified to stay inert when registryCorpusEmbeddingEnabled() is false.

src/lib/rag-extractive-answer.ts (1)

287-299: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Query-side blood-count regex is narrower than the text-side evidence regex.

requiresBloodCountEvidence only matches anc|fbc|wbc|wcc|neutrophil(s) in the query, while hasBloodCountEvidence (used to validate the evidence text) additionally recognizes full blood count, white blood cell, white cell, and blood count. A query phrased as "What full blood count threshold should withhold clozapine?" would fail to trigger the new gate at Line 321/584 even though it's clearly a blood-count query, silently falling back to the looser pre-existing checks. Consider aligning the query-detection regex with the same term list used for text evidence.

♻️ Suggested fix to align term lists

 function requiresBloodCountEvidence(query: string) {
-  return /\b(?:anc|fbc|wbc|wcc|neutrophil|neutrophils)\b/i.test(query);
+  return /\b(?:anc|fbc|full blood count|wbc|wcc|white blood cell|white cell|neutrophil|neutrophils|blood count)\b/i.test(
+    query,
+  );
 }

Also applies to: 321-322, 584-585

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/rag-extractive-answer.ts` around lines 287 - 299, The blood-count
query gate in requiresBloodCountEvidence is too narrow compared with
hasBloodCountEvidence, so phrases like “full blood count” or “white blood cell”
won’t trigger the stricter path. Update requiresBloodCountEvidence in
rag-extractive-answer.ts to use the same term set as hasBloodCountEvidence (or
share a common pattern/helper) so the query-side detection matches the
evidence-side detection. Keep the existing withhold-action flow intact where the
gate is used.

src/lib/rag.ts (1)

2666-2685: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚖️ Poor tradeoff

hasAdmissionCommunityLookupIntent is duplicated across modules with a different normalizer.

An identically named hasAdmissionCommunityLookupIntent exists in src/lib/rag-routing.ts (line 281) using normalizeLookupText, whereas this copy uses normalizeDocumentAliasText. Divergent normalization behind the same name/intent risks the routing gate and the fast-path/coverage gate disagreeing on what counts as an admission-community query. Consider sharing a single helper (or at least aligning the normalizer) to keep the two gates consistent.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/rag.ts` around lines 2666 - 2685, The admission-community intent
helper is duplicated with different normalization logic, so the routing gate and
coverage gate can disagree. Update the `hasAdmissionCommunityLookupIntent`
implementation in `rag.ts` to use the same normalization approach as the copy in
`rag-routing.ts` (or extract a shared helper used by both), and keep the
title-support checks in `hasAdmissionCommunityTitleSupport` aligned with that
shared intent logic.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/branch-cleanup-guide.md`:
- Around line 44-53: The ledger lookup in the branch review filter interpolates
$branch directly into the Select-String pattern, so branch names with regex
metacharacters can match the wrong row. Update the lookup in the branch-cleanup
guide to treat the branch name as a literal when building the pattern, while
keeping the existing checks against $branch, $head, and the branch-cleanup scope
in the review-ledger lookup.

In `@scripts/embed-registry-records.ts`:
- Around line 123-161: The owner-count aggregation in listEligibleOwnerCounts
currently uses plain Supabase selects for clinical_registry_records,
medication_records, and differential_records, so results can be truncated by Max
Rows. Update this function to page through each table with .range() or switch to
a grouped/count query so the counts remain exact, and keep the existing
ensureOwnerCount and sort logic intact.

In `@src/lib/rag-routing.ts`:
- Around line 244-249: The querySpecificTitleTokens helper is stemming tokens
before checking specificTitleStopWords, which lets stopwords like “does” turn
into junk tokens such as “doe” and affect hasSpecificTitleSupport scoring.
Update querySpecificTitleTokens so the stopword filter runs on the normalized
token before the trailing-`s` replacement, then apply stemming only to tokens
that are not in specificTitleStopWords and still pass the length check.

In `@src/lib/registry-corpus.ts`:
- Around line 289-324: The batch write in registryCorpus currently rolls back
only newly inserted documents, but it leaves pre-existing documents overwritten
by the documents upsert if the subsequent document_chunks upsert fails. Update
the batch flow in registryCorpus to make document and chunk writes atomic:
either snapshot and restore the affected document rows before retrying/throwing,
or move the paired write logic into a single transactional Postgres RPC so both
the documents and document_chunks changes succeed or fail together. Ensure the
rollback path covers both newly inserted and pre-existing rows touched by the
batch.

---

Nitpick comments:
In `@src/lib/rag-extractive-answer.ts`:
- Around line 287-299: The blood-count query gate in requiresBloodCountEvidence
is too narrow compared with hasBloodCountEvidence, so phrases like “full blood
count” or “white blood cell” won’t trigger the stricter path. Update
requiresBloodCountEvidence in rag-extractive-answer.ts to use the same term set
as hasBloodCountEvidence (or share a common pattern/helper) so the query-side
detection matches the evidence-side detection. Keep the existing withhold-action
flow intact where the gate is used.

In `@src/lib/rag.ts`:
- Around line 2666-2685: The admission-community intent helper is duplicated
with different normalization logic, so the routing gate and coverage gate can
disagree. Update the `hasAdmissionCommunityLookupIntent` implementation in
`rag.ts` to use the same normalization approach as the copy in `rag-routing.ts`
(or extract a shared helper used by both), and keep the title-support checks in
`hasAdmissionCommunityTitleSupport` aligned with that shared intent logic.

In `@src/lib/registry-corpus.ts`:
- Around line 431-448: The fallback result in
bestEffortReembedRegistryRecordAfterEdit duplicates the skipped
RegistryCorpusEmbedResult shape instead of reusing the shared helper. Update the
catch path to construct the failure/skip response through
skippedRegistryEmbedResult (the helper used in this module) so the result shape
stays centralized and consistent, while still preserving the errorMessage from
embedError and the existing console.error context in
bestEffortReembedRegistryRecordAfterEdit.

In `@tests/registry-corpus.test.ts`:
- Around line 74-98: The disabled-embedding test in registry-corpus only
exercises the reembedRegistryRecordAfterEdit medication path, so it can miss
regressions in other dispatch branches. Extend the existing test around
reembedRegistryRecordAfterEdit to parameterize over the service and differential
flows as well, covering reembedClinicalRegistryRecordBySlug and
reembedDifferentialRecordBySlug so each branch is verified to stay inert when
registryCorpusEmbeddingEnabled() is false.

🪄 Autofix (Beta)

✅ Autofix completed

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3171f0ce-4c7c-470a-a8f9-1ddcadf8aa72

📥 Commits

Reviewing files that changed from the base of the PR and between ac5a6e1 and f55378d.

📒 Files selected for processing (35)

• .cursor/skills/accessibility-review/SKILL.md
• .cursor/skills/ai-architecture-review/SKILL.md
• .cursor/skills/api-review/SKILL.md
• .cursor/skills/code-quality-review/SKILL.md
• .cursor/skills/design-review/SKILL.md
• .cursor/skills/frontend-architecture-review/SKILL.md
• .cursor/skills/performance-review/SKILL.md
• .cursor/skills/release-readiness-review/SKILL.md
• .cursor/skills/repo-auditor/SKILL.md
• .cursor/skills/security-review/SKILL.md
• .cursor/skills/supabase-postgres-best-practices/SKILL.md
• .cursor/skills/testing-review/SKILL.md
• .cursor/skills/ux-review/SKILL.md
• .github/workflows/codex-autofix-review-comments.yml
• .github/workflows/live-drift.yml
• AGENTS.md
• docs/branch-cleanup-guide.md
• docs/branch-review-ledger.md
• docs/codex-prompt-playbook.md
• docs/codex-review-protocol.md
• docs/rag-hybrid-findings-and-todo.md
• scripts/embed-registry-records.ts
• src/lib/clinical-search.ts
• src/lib/rag-extractive-answer.ts
• src/lib/rag-retrieval-variants.ts
• src/lib/rag-routing.ts
• src/lib/rag.ts
• src/lib/registry-corpus.ts
• supabase/drift-manifest.json
• supabase/schema.sql
• tests/clinical-search.test.ts
• tests/rag-answer-fallback.test.ts
• tests/rag-routing.test.ts
• tests/registry-corpus.test.ts
• tests/retrieval-query-variants.test.ts

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merge conflicts resolved in commit cb5164a. The conflicts were in:

• .github/workflows/codex-autofix-review-comments.yml - accepted the throttled workflow version from main
• AGENTS.md - accepted the updated documentation from main

Both files now use the improved versions from the main branch.

[coderabbitai]

Note

Autofix is a beta feature. Expect some limitations and changes as we gather feedback and continue to improve it.

Fixes Applied Successfully

Fixed 4 file(s) based on 4 unresolved review comments.

Files modified:

• docs/branch-cleanup-guide.md
• scripts/embed-registry-records.ts
• src/lib/rag-routing.ts
• src/lib/registry-corpus.ts

Commit: e681ae3312b69fe3e20e8e3e16ca4f328ca6dadd

The changes have been pushed to the codex/branch-cleanup-main-merge branch.

Time taken: 6m 19s