Fix search page defects across documents, services, forms, favourites, and differentials

[BigSimmo]

Summary

• Reviewed every search surface (root dashboard modes, /documents/search, /services, /forms, /favourites, /differentials) and fixed the confirmed defects:
• URL builder (src/lib/app-modes.ts): appModeHomeHref built malformed double-? URLs for prescribing/tools mode homes (e.g. /?mode=prescribing?focus=1), which failed mode parsing and dropped users into Answer mode. It now uses the right separator, and namespaced modes only emit run=1 alongside a query.
• Documents search (master-document-flow-mockups.tsx): the results filter OR-ed the query, type, and default-slug predicates, so the query never filtered anything and the type tabs could only ever add results — now query AND type, with an explicit no-results state. Result rows surface evidence matching the active type tab (instead of always evidence[0] with a hardcoded table icon), documents with no extracted evidence no longer borrow the clozapine fixture's evidence in search rows, reader evidence tab counts are derived from the document rather than hardcoded to 1, "Open document" from evidence context preserves the viewed evidence page/chunk, and the evidence sidebar shows the document's own source path and updated date.
• Services (services-navigator-page.tsx, services/page.tsx): the navigator no longer flashes a false "No matches" while the registry loads (skeleton + loading header band); the "ATSI-specific" quick chip now searches "Aboriginal Torres Strait Islander" (the term the catalogue actually contains — atsi appears nowhere and the chip previously matched on the word "specific"); the server wrapper normalizes an array-valued run param the same way it already normalized q/query.
• Forms (forms-search-results-page.tsx): the empty state now offers a working "try an example" action (mirrors services); result badge codes map to the form record by slug instead of row position, so ranking changes can no longer disagree with the pathway panel (4A = Transport order etc.); the "Matched because" column interpolates the actual query instead of a hardcoded "transport".
• Favourites (favourites-library-nav.tsx, favourites-command-library-page.tsx): selecting a Quick View clears a lingering set filter whose sidebar highlight would otherwise disappear while still filtering results; the sort control is disabled in the Recently used view it cannot affect; the continue strip is gated on scope-filtered items so it can't advertise an item absent from the list below; run-action saved searches are labelled "Saved search" instead of "Source".
• Differentials (src/lib/differentials.ts): snapshot loaders now drop markdown-header preset artifacts (the literal # Scenario Presets pill in "Recent work") and numeric field-weight alias entries (e.g. tags → ["1.1"]) that made /api/differentials query expansion match unrelated records by number substrings.

Review findings deliberately not changed: the forms ranker's services guard and injected catalogue tokens are intentional (asserted by tests/forms.test.ts), and /differentials?q=…&run=1 / /forms?q=…&run=1 already render working results via the GlobalSearchShell layout override.

Verification

• npm run verify:cheap
• npm run verify:ui when UI, routing, styling, browser behavior, reduced-motion, or forced-colors behavior changed — not run: this environment has no configured browser project secrets; npm run typecheck, lint, and the full vitest suite (1097 passing) were run instead
• npm run verify:release before release or handoff confidence claims — not applicable to this PR
• npm run format:check
• npm run eval:retrieval:quality (must stay 23/23) — not run: requires live Supabase/OpenAI keys unavailable in this environment. RAG retrieval/ranking/chunking/scoring code paths are untouched; the only search-logic changes are to client-side fixture filtering (documents mockup) and the differentials static-catalog alias map (which previously matched records on leaked numeric weights)
• npm run eval:rag / eval:quality — answer generation and synthesis are untouched
• npm run check:production-readiness — no clinical workflow, privacy, environment, Supabase, or deployment behavior changed
• npm run check:deployment-readiness — no deployment behavior changed

Clinical Governance Preflight

• Source-backed claims still require linked source verification before clinical use
• No patient-identifiable document workflow was introduced or expanded without explicit governance approval
• Supabase target remains Clinical KB Database (sjrfecxgysukkwxsowpy) — no env or Supabase changes
• Service-role keys and private document access remain server-only
• Demo/synthetic content remains clearly separated from real clinical sources (documents search rows now stop borrowing another document's fixture evidence, improving this separation)
• Source metadata, review status, and outdated/unknown-source behavior remain conservative
• Deployment classification/TGA SaMD impact was checked when clinical decision-support behavior changed — no clinical decision-support behavior changed

Notes

• The documents /documents/search surface is fixture-backed by design; the filter fix makes the sample behave like a real search (query narrows, tabs narrow, empty state reachable).
• The differentials snapshot itself was left untouched; junk rows are filtered at the loader so regeneration cannot reintroduce the bug silently.

🤖 Generated with Claude Code

https://claude.ai/code/session_015rc3tTjs65D4gFBuCQhZGs

---

Generated by Claude Code

---

Note

Medium Risk
Touches shared URL building and several user-facing search flows (including differentials query expansion), but changes are defensive fixes to filtering and routing rather than new clinical logic or backend retrieval.

Overview
Fixes search and navigation defects across multiple clinical surfaces so queries, filters, and deep links behave consistently.

appModeHomeHref now uses & when appending params to mode homes that already contain ?, and only adds run=1 when a query is present—avoiding broken URLs and spurious “submitted search” states.

Documents search mockup tightens filtering to query AND evidence type (replacing OR logic that ignored the query), adds empty states, and aligns rows with the active evidence tab. Documents without extracted evidence show snippets and “No extracted evidence” instead of borrowing another fixture; reader/evidence links preserve page/chunk; evidence tab counts and sidebar metadata come from the active document.

Services shows a loading skeleton and header loading state while the registry loads (no premature empty state), normalizes array run search params on the server route, and maps the ATSI quick chip to catalogue-friendly search text.

Forms wires empty-state “try an example” through appModeHomeHref, maps pathway badge codes by form slug, and surfaces the user’s query in “Matched because” copy.

Favourites clears set filters when switching Quick views, disables sort in Recently used, gates the continue strip on scope-filtered items, and labels run-action source items as Saved search.

Differentials loader filters markdown-header preset junk and numeric-only search aliases from generated snapshots so API search expansion does not match unrelated records.

^{Reviewed by Cursor Bugbot for commit 4a31942. 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.