Add cross-mode entity quick links to answer search results

[BigSimmo]

Summary

• When a search names an entity that lives in another app mode — a medication ("what is the maximum dose of clozapine?"), a service, a form, or a differential — an "Also in your library" card strip renders with the results. Each card shows the mode icon, entity title, a one-line context, up to two badges, and a mode pill; the title links to the entity's detail page (e.g. /medications/clozapine) and a secondary button re-runs the search inside that mode.
• Surfaces: under Answer-mode results (thread-aware — follow-ups like "what about renal impairment?" keep the entity's card by walking back through prior turns), on dashboard documents results, and on the /documents/search command centre. All three mount a self-contained CrossModeLinksSection that owns catalog fetching + matching.
• Matching: pure matcher src/lib/cross-mode-links.ts runs the shared rankCatalogRecords rankers with name/title precision gates; results capped 2 per mode / 4 total, deduped across the services/forms registries.
• Ranker precision fix (all catalog search surfaces): field matching in rankCatalogRecords is now word-boundary aligned, so query terms cannot hide inside entity names ("renal" no longer name-matches "adrenaline"); prefix matching preserves search-as-you-type.
• Slim payload: /api/medications?fields=index returns the ~30 KB identity slice instead of the ~3.4 MB catalog; the links surface uses it. Differentials snapshot stays code-split behind a dynamic import.
• Click telemetry: /api/search/interaction accepts a crossMode target and stores the same privacy-hardened rag_query_misses row with the mode/slug in metadata (interaction: "cross_mode_link_open"), so retrieval reviews can see whether the strip earns its space.
• Answer crossModes command-surface config now lists services and forms, matching the strip's coverage. keywordQueryFromNaturalLanguage hoisted to src/lib/keyword-query.ts (re-exported for existing consumers). docs/process-hardening.md records this workstream's verification state.

Verification

• npm run verify:cheap equivalents — lint 0 errors, typecheck clean, 1157 vitest tests pass (incl. new tests/cross-mode-links.test.ts, ranker word-boundary regression in tests/medications.test.ts, fields=index case in tests/medications-route.test.ts, cross-mode telemetry cases in tests/search-interaction-route.test.ts)
• npm run verify:ui — full Chromium suite: 107 passed, 0 failed (8.8m) on the follow-up work; the earlier CI ui-smoke failure ("prescribing mode home top-aligns on phones") reproduces as flaky only on the CI runner and passes locally on the same commit
• Manual browser proof: answer-mode strip (Medication + Differentials cards for a clozapine question, click-through to /medications/clozapine) and /documents/search strip (3 cards above the results table) verified against the live dev server in demo mode
• npm run verify:release — not run (needs live keys; debt recorded in docs/process-hardening.md)
• npm run check:production-readiness — fails only on missing Supabase/OpenAI secrets, documented as expected in this environment (AGENTS.md)
• npm run eval:retrieval:quality / eval:rag — not run here (no live keys). Note: the word-boundary ranker change touches catalog search (medications/services/forms/tools/differentials), not the RAG document-retrieval pipeline; document retrieval scoring is unchanged.

Clinical Governance Preflight

• Source-backed claims still require linked source verification before clinical use — the strip links only to existing governed record pages; no new clinical content is generated
• No patient-identifiable document workflow was introduced or expanded
• Query privacy preserved — cross-mode telemetry reuses queryTextForStorage/queryPrivacyMetadata redaction; no raw query text persisted beyond the existing policy; unit tests pin the redaction
• Supabase target remains Clinical KB Database (sjrfecxgysukkwxsowpy) — no Supabase config touched
• Service-role keys and private document access remain server-only
• Demo/synthetic content remains clearly separated from real clinical sources
• Deployment classification/TGA SaMD impact was checked — navigational surface only; no clinical decision-support behavior changed

Notes

• The Playwright coverage lives in tests/ui-smoke.spec.ts because playwright.config.ts testMatch only picks up the fixed spec set; a new file would silently never run.
• Cross-mode telemetry rows have clicked_document_id: null with the target in metadata — miss aggregations by document should filter on metadata.interaction.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Nt7UBgqg82NEPinWUZ6PXF

[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 default effort and found 1 potential issue.

Bugbot Autofix prepared a fix for the issue found in the latest run.

• ✅ Fixed: Cross-mode links drop after second follow-up
  • The crossModeLinks memo now walks prior answer turns from newest to oldest when the current query yields no matches, so entity cards persist across multi-hop follow-ups.

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

^{Reviewed by Cursor Bugbot for commit 954daa2. Configure here.}

[BigSimmo]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merge conflicts resolved. There was a single conflict in tests/medications.test.ts where both branches added a new test for rankMedicationRecords — I kept both tests:

• Ours: "does not treat mid-word substrings as name matches" (word-boundary precision fix)
• main's: "boosts name-prefix matches above content-only matches" (name-prefix boost)

All 5 medication tests pass. Commit: db2b21a