Wire differentials, services and medications catalogues into search modes

[BigSimmo]

Summary

• Differentials search mode now searches the imported catalogue (201 diagnoses + 31 presentations). Previously the results panel rendered a hard-coded "acute confusion" demo fixture regardless of the query. src/lib/differentials.ts gains scored, alias-aware rankers (rankDifferentialRecords, rankPresentationWorkflows) and an adaptive composer (composeDifferentialSearchResults) that leads with a presentation workflow when it matches about as strongly as the best diagnosis (followed by its candidate diagnoses in ranked order), otherwise interleaves by score. Urgency (emergent/urgent) only nudges tie-breaks — it never outranks a stronger text match.
• GET /api/differentials returns scored matches ({record|workflow, score, reasons}) for both kinds, mirroring the medications API, and ranks live owner rows directly instead of the previous O(n²) snapshot-intersect. records/presentations/total/governance keep their existing semantics (additive change).
• New useDifferentialSearch(query) hook fetches diagnosis + presentation matches in parallel with the sibling hooks' auth/401 handling; the Differentials results view now renders real query-driven results with loading/empty/error states, functional All/Presentations/Diagnoses filters with real counts, real tag-overflow chips (no more fake +2), and accurate catalogue copy (the "synthetic demonstration content" banner is replaced by differentials-catalogue-notice).
• Dashboard wiring: the results view is gated on search submission (persists with zero document evidence), and an empty document-corpus result in differentials mode applies empty evidence instead of erroring — catalogue results are the primary content. The /api/search document-evidence call is kept and runs in parallel.
• Medications (prescribing) polish: the hard-coded acamprosate row highlight and alcohol/acamprosate/baclofen filter literals are replaced with rank-driven equivalents, and rankMedicationRecords gains a +5 name-prefix boost (reason "name prefix").
• Services: already wired to the imported 219-service catalogue with scored client-side ranking — no change.

Verification

• npm run verify:cheap — pass (lint, typecheck, 1108 vitest tests; only pre-existing warnings)
• npm run verify:ui — targeted equivalent: full tests/ui-tools.spec.ts on chromium against the live dev server, 36/36 passed (includes the rewritten differentials search spec asserting a real catalogue result row); prescribing smoke spec also passed
• npm run verify:release — not run (not a release/handoff change)
• npm run format:check — pass
• npm run eval:retrieval:quality — not run: needs live keys unavailable in this environment. Note: this change does not modify document-retrieval ranking/selection/chunking/scoring (src/lib/rag.ts, retrieval-selection.ts untouched); it adds catalogue-side rankers and client wiring only. Please run locally before merge if you want the 23/23 confirmation.
• npm run eval:rag / eval:quality — not applicable (answer generation untouched)
• npm run check:production-readiness — runs clean apart from expected missing-secret findings in this no-secrets environment (no Supabase/OpenAI env configured)
• npm run check:deployment-readiness — not applicable

Clinical Governance Preflight

• Source-backed claims still require linked source verification before clinical use — the document-evidence panel and "Run source search" flows are preserved; catalogue results are labelled as imported catalogue content with source counts from the indexed library
• No patient-identifiable document workflow was introduced or expanded
• Supabase target remains Clinical KB Database (sjrfecxgysukkwxsowpy) — no env/project changes
• Service-role keys and private document access remain server-only — API routes unchanged in their access model
• Demo/synthetic content remains clearly separated from real clinical sources — this PR removes a synthetic demo fixture from the search flow and replaces it with the imported, locally reviewed catalogue; catalogue copy states the provenance
• Source metadata, review status, and outdated/unknown-source behaviour remain conservative — governance payloads untouched
• Deployment classification/TGA SaMD impact checked: ranking of the existing curated decision-support catalogue changed from static demo to query-relevance ordering; the decision-support disclaimers and review-before-use notices remain in place

Notes

• Ranker scoring mirrors the existing rankMedicationRecords/rankServiceRecords pattern (field-weighted term matching with reasons), plus catalogue alias expansion (e.g. "confused" → confusion/delirium/encephalopathy) surfaced as a symptom alias reason.
• Follow-ups deliberately out of scope: query-filtering the /differentials/presentations and /differentials/diagnoses stream pages (still static cards), and debouncing the per-keystroke catalogue fetches (matches the existing medications behaviour).

🤖 Generated with Claude Code

https://claude.ai/code/session_015UuyyMMegXxTeEJsyR741t

---

Generated by Claude Code

---

Note

Medium Risk
Changes clinical decision-support result ordering and copy from static demo to relevance-ranked catalogue content; document-evidence flows are preserved but users may rely on ranked catalogue rows before source search.

Overview
Differentials search replaces the hard-coded acute-confusion demo with query-driven results from the imported catalogue (diagnoses and presentations). New rankDifferentialRecords / rankPresentationWorkflows scoring (alias expansion, urgency tie-breaks) and composeDifferentialSearchResults adapt ordering (presentation-led when it matches strongly). GET /api/differentials now returns optional matches with score and reasons, ranking owner rows in one pass instead of intersecting snapshot search.

The UI adds useDifferentialSearch, real loading/empty/error states, All / Presentations / Diagnoses filters, and catalogue copy instead of demonstration banners. ClinicalDashboard shows ranked results after submit even with zero document evidence, and treats an empty corpus in differentials mode as empty evidence rather than a blocking error.

Prescribing drops hard-coded medication highlights/filters in favor of rank-based selection and adds a name-prefix boost to rankMedicationRecords. Tests and UI specs updated accordingly.

^{Reviewed by Cursor Bugbot for commit eda08d6. Configure here.}

[supabase]

This pull request has been ignored for the connected project sjrfecxgysukkwxsowpy because there are no changes detected in supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.

---

Preview Branches by Supabase.
Learn more about Supabase Branching ↗︎.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 3 potential issues.

Bugbot Autofix prepared fixes for all 3 issues found in the latest run.

• ✅ Fixed: Stale source evidence after query edit
  • Source evidence now tracks the query that produced documentMatches and is hidden when the live composer text no longer matches.
• ✅ Fixed: Filter treats first row as best
  • Result rows now use the global best-match id and catalogue rank instead of the filtered list index.
• ✅ Fixed: Catalogue failure shows empty-match UI
  • The empty-results state is skipped when catalogue search fails so only the failure alert is shown.

_{You can send follow-ups to the cloud agent here.}

^{Reviewed by Cursor Bugbot for commit eda08d6. Configure here.}