Finish RAG registry re-embed and quality routing#438
Merged
BigSimmo merged 4 commits intoJul 9, 2026
Conversation
Contributor
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughThis PR adjusts RAG query routing and retrieval to favor extractive answers for medication/title-supported and clozapine blood-withhold queries, tightens red-result-action evidence gating, adds admission/community query handling, introduces batched registry-corpus embedding with re-embed-on-edit support, adds a CLI owner-listing mode, and updates schema permissions, drift manifest, and docs/tests accordingly. ChangesRAG Query Routing and Extractive Gating
Estimated code review effort: 4 (Complex) | ~60 minutes Registry Corpus Embedding and Re-embed Support
Estimated code review effort: 3 (Moderate) | ~25 minutes Sequence Diagram(s)sequenceDiagram
participant Client
participant chooseAnswerRoute
participant hasSpecificTitleSupport
participant shouldUseExtractiveMedicationLookup
participant ExtractiveAnswer
Client->>chooseAnswerRoute: submit query + retrieval results
chooseAnswerRoute->>hasSpecificTitleSupport: check title/token overlap
hasSpecificTitleSupport-->>chooseAnswerRoute: specificTitleSupport result
chooseAnswerRoute->>shouldUseExtractiveMedicationLookup: evaluate medication/threshold intent
shouldUseExtractiveMedicationLookup-->>chooseAnswerRoute: extractive eligibility
chooseAnswerRoute-->>Client: route = extractive/strong/fast
Client->>ExtractiveAnswer: generate answer (if extractive)
ExtractiveAnswer->>ExtractiveAnswer: check blood-count/withhold-action evidence
ExtractiveAnswer-->>Client: gated answer or fail-closed
sequenceDiagram
participant Operator
participant EmbedScript
participant Supabase
participant RegistryCorpus
Operator->>EmbedScript: run --list-owners
EmbedScript->>Supabase: query row counts per owner_id
Supabase-->>EmbedScript: counts across record tables
EmbedScript-->>Operator: print eligible owner totals
Operator->>EmbedScript: run --write --owner-id
EmbedScript->>RegistryCorpus: embedRegistryCorpusEntries (batched)
RegistryCorpus->>Supabase: upsert documents/document_chunks per batch
Supabase-->>RegistryCorpus: batch upsert result
RegistryCorpus-->>EmbedScript: total documentCount/chunkCount
Caution Pre-merge checks failedPlease resolve all errors before merging. Addressing warnings is optional.
❌ Failed checks (1 error, 2 warnings)
✅ Passed checks (8 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
✨ Simplify code
Comment |
|
Updates to Preview Branch (codex/rag-registry-reembed-routing) ↗︎
Tasks are run on every commit but only new migration files are pushed.
View logs for this Workflow Run ↗︎. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7c9e8e0ad4
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".
Owner
Author
|
@copilot resolve the merge conflicts in this pull request |
Contributor
Resolved in commit Three conflicts were resolved:
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/lib/rag-routing.ts (1)
587-608: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick winKeep high-confidence comparison queries on the synthesis path.
src/lib/rag-extractive-answer.tsonly groups extracted facts by kind and never produces acomparisonsection, so routingqueryClass === "comparison"toextractivecan return a flat fact list instead of a source-by-source reconciliation. Keepstronghere unless a dedicated comparison extractor is added.🤖 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-routing.ts` around lines 587 - 608, The comparison routing block in rag-routing should not send queryClass === "comparison" to the extractive path, since rag-extractive-answer only groups facts and does not generate a comparison section. Update the decision in the comparison handling branch to keep these queries on the strong synthesis path, and only allow extractive for the high-confidence case when it is appropriate for non-comparison retrieval.
🧹 Nitpick comments (3)
scripts/embed-registry-records.ts (2)
106-161: 🚀 Performance & Scalability | 🔵 Trivial | 💤 Low valueOwner-count queries fetch entire tables client-side to compute counts.
listEligibleOwnerCountspulls every row'sowner_id/kindfromclinical_registry_records,medication_records, anddifferential_recordswith no pagination, then aggregates in memory. Fine for an admin CLI at current registry sizes, but if these tables grow substantially this becomes an unbounded full-table read on every--list-ownersinvocation.🤖 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 `@scripts/embed-registry-records.ts` around lines 106 - 161, The listEligibleOwnerCounts helper is doing full-table client-side aggregation across clinical_registry_records, medication_records, and differential_records, which scales poorly. Update this flow to avoid reading every row into memory: either compute counts with server-side aggregation/grouping in Supabase or add paginated/limited batching inside listEligibleOwnerCounts, keeping the same ensureOwnerCount and sorting behavior for the final OwnerCounts result.
216-220: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueZero-total guard only covers the dry-run path.
The new "no eligible rows" message only prints under
!args.write. If--write --confirmis run against an owner with zero eligible entries, the script still prompts "Embed and upsert 0 registry corpus entries?" — harmless (the downstreamembedRegistryCorpusEntriesshort-circuits on empty input) but a confusing UX gap given the new--list-ownersguidance is meant to prevent exactly this confusion.🤖 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 `@scripts/embed-registry-records.ts` around lines 216 - 220, The zero-eligible-rows guard in the registry embed flow only reports the “no eligible rows” case during the non-write path, so `--write --confirm` still reaches the confirmation prompt with a count of 0. Update the logic around the `total === 0` check in `embed-registry-records.ts` so the `console.log` guidance runs before any write/confirm branching, preventing the `embedRegistryCorpusEntries` prompt from appearing for empty input and keeping the `--list-owners` hint consistent.src/lib/registry-corpus.ts (1)
269-293: 🚀 Performance & Scalability | 🔵 Trivial | 💤 Low valueBatching looks correct; partial-failure counts are lost on throw.
If a later batch's upsert fails, the counts from previously-committed batches are discarded (the function throws before returning). Since
upsertis idempotent onid, a retry is safe, but callers get no visibility into how much succeeded before the failure — worth considering for observability/logging on retry-heavy embedding runs.🤖 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 269 - 293, embedRegistryCorpusEntries currently throws on a later batch failure before callers can see how many documents/chunks were already committed. Update the error handling in embedRegistryCorpusEntries (and the document/chunk upsert blocks) to preserve or log the running documentCount and chunkCount when a batch fails, so retry-heavy runs have visibility into partial progress even though the operation aborts.
🤖 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 `@src/lib/registry-corpus.ts`:
- Around line 391-431: The re-embed helper in reembedRegistryRecordAfterEdit is
currently disconnected from the registry update flow, so edits never trigger
embedding refreshes. Hook bestEffortReembedRegistryRecordAfterEdit into the
registry write/edit path in src/app/api/registry/records, using the existing
RegistryCorpusEditTarget data after a successful write so the correct re-embed
branch runs for medication, differential, or clinical records. Make sure the
call happens only on mutation success and preserves the existing scope/error
handling behavior.
---
Outside diff comments:
In `@src/lib/rag-routing.ts`:
- Around line 587-608: The comparison routing block in rag-routing should not
send queryClass === "comparison" to the extractive path, since
rag-extractive-answer only groups facts and does not generate a comparison
section. Update the decision in the comparison handling branch to keep these
queries on the strong synthesis path, and only allow extractive for the
high-confidence case when it is appropriate for non-comparison retrieval.
---
Nitpick comments:
In `@scripts/embed-registry-records.ts`:
- Around line 106-161: The listEligibleOwnerCounts helper is doing full-table
client-side aggregation across clinical_registry_records, medication_records,
and differential_records, which scales poorly. Update this flow to avoid reading
every row into memory: either compute counts with server-side
aggregation/grouping in Supabase or add paginated/limited batching inside
listEligibleOwnerCounts, keeping the same ensureOwnerCount and sorting behavior
for the final OwnerCounts result.
- Around line 216-220: The zero-eligible-rows guard in the registry embed flow
only reports the “no eligible rows” case during the non-write path, so `--write
--confirm` still reaches the confirmation prompt with a count of 0. Update the
logic around the `total === 0` check in `embed-registry-records.ts` so the
`console.log` guidance runs before any write/confirm branching, preventing the
`embedRegistryCorpusEntries` prompt from appearing for empty input and keeping
the `--list-owners` hint consistent.
In `@src/lib/registry-corpus.ts`:
- Around line 269-293: embedRegistryCorpusEntries currently throws on a later
batch failure before callers can see how many documents/chunks were already
committed. Update the error handling in embedRegistryCorpusEntries (and the
document/chunk upsert blocks) to preserve or log the running documentCount and
chunkCount when a batch fails, so retry-heavy runs have visibility into partial
progress even though the operation aborts.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: eefecae9-52a3-434f-80ec-864f781ff0c2
📒 Files selected for processing (15)
docs/rag-hybrid-findings-and-todo.mdscripts/embed-registry-records.tssrc/lib/clinical-search.tssrc/lib/rag-extractive-answer.tssrc/lib/rag-retrieval-variants.tssrc/lib/rag-routing.tssrc/lib/rag.tssrc/lib/registry-corpus.tssupabase/drift-manifest.jsonsupabase/schema.sqltests/clinical-search.test.tstests/rag-answer-fallback.test.tstests/rag-routing.test.tstests/registry-corpus.test.tstests/retrieval-query-variants.test.ts
Comment on lines
+391
to
+431
| export function reembedRegistryRecordAfterEdit( | ||
| supabase: AdminClient, | ||
| target: RegistryCorpusEditTarget, | ||
| ): Promise<RegistryCorpusEmbedResult> { | ||
| if (target.corpusKind === "medication") { | ||
| return reembedMedicationRecordBySlug(supabase, { ownerId: target.ownerId, slug: target.slug }); | ||
| } | ||
|
|
||
| if (target.corpusKind === "differential") { | ||
| return reembedDifferentialRecordBySlug(supabase, { | ||
| ownerId: target.ownerId, | ||
| slug: target.slug, | ||
| kind: target.differentialKind, | ||
| }); | ||
| } | ||
|
|
||
| return reembedClinicalRegistryRecordBySlug(supabase, { | ||
| ownerId: target.ownerId, | ||
| slug: target.slug, | ||
| kind: target.corpusKind, | ||
| }); | ||
| } | ||
|
|
||
| export async function bestEffortReembedRegistryRecordAfterEdit(args: { | ||
| supabase: AdminClient; | ||
| target: RegistryCorpusEditTarget; | ||
| scope: string; | ||
| }) { | ||
| if (!registryCorpusEmbeddingEnabled()) return skippedRegistryEmbedResult("disabled"); | ||
|
|
||
| try { | ||
| return await reembedRegistryRecordAfterEdit(args.supabase, args.target); | ||
| } catch (embedError) { | ||
| const errorMessage = embedError instanceof Error ? embedError.message : String(embedError); | ||
| console.error( | ||
| `[${args.scope}] corpus re-embedding failed after registry edit for ${args.target.ownerId}/${args.target.corpusKind}/${args.target.slug}`, | ||
| embedError, | ||
| ); | ||
| return { documentCount: 0, chunkCount: 0, skipped: true, reason: "failed", errorMessage } satisfies RegistryCorpusEmbedResult; | ||
| } | ||
| } |
Contributor
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Check whether the new re-embed-on-edit functions are actually called anywhere (edit routes/actions)
rg -n -C3 'reembedRegistryRecordAfterEdit|bestEffortReembedRegistryRecordAfterEdit' --type=ts -g '!**/registry-corpus.ts' -g '!**/registry-corpus.test.ts'Repository: BigSimmo/Database
Length of output: 155
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '\n== direct references ==\n'
rg -n -C2 'reembedRegistryRecordAfterEdit|bestEffortReembedRegistryRecordAfterEdit|RegistryCorpusEditTarget' --type=ts
printf '\n== likely registry edit entrypoints ==\n'
rg -n -C2 'registry edit|edit registry|update registry|registry.*update|update.*registry|corpusKind|ownerId.*slug' --type=ts srcRepository: BigSimmo/Database
Length of output: 180
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '\n== files mentioning registry edit/update ==\n'
rg -n --files src docs | rg 'registry|corpus|edit|update'
printf '\n== registry corpus edit target and edit-related calls ==\n'
rg -n -C3 'RegistryCorpusEditTarget|bestEffortReembedRegistryRecordAfterEdit|reembedRegistryRecordAfterEdit|reembed.*AfterEdit|afterEdit|onEdit|update.*corpus|corpus.*update' src docs
printf '\n== API route / action candidates ==\n'
rg -n -C3 'export (async )?function|export const (GET|POST|PUT|PATCH|DELETE)|"use server"|server action|route' src/app src/lib 2>/dev/null || trueRepository: BigSimmo/Database
Length of output: 50374
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '\n== candidate route/action files ==\n'
rg --files src/app src | rg '(registry|corpus|medication|differential|edit|update|route)\.(ts|tsx|js|jsx)$' | sort
printf '\n== direct re-embed references in candidate files ==\n'
rg -n -C3 'reembedRegistryRecordAfterEdit|bestEffortReembedRegistryRecordAfterEdit|reembed.*AfterEdit|afterEdit' \
$(rg --files src/app src | rg '(registry|corpus|medication|differential|edit|update|route)\.(ts|tsx|js|jsx)$' | tr '\n' ' ') 2>/dev/null || trueRepository: BigSimmo/Database
Length of output: 4223
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '\n== registry edit routes ==\n'
for f in src/app/api/registry/records/route.ts src/app/api/registry/records/[slug]/route.ts; do
echo "--- $f"
sed -n '1,260p' "$f"
done
printf '\n== docs item 22 ==\n'
rg -n -C3 'item 22|re-embed-on-edit hooks for registry updates|re-embed-on-edit' docs/rag-hybrid-findings-and-todo.mdRepository: BigSimmo/Database
Length of output: 11632
Wire the re-embed hook into the registry write path
This helper is still unused: the registry routes under src/app/api/registry/records are GET-only, and there are no callers elsewhere. Until the edit path invokes it, registry updates will leave embeddings stale.
🤖 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 391 - 431, The re-embed helper in
reembedRegistryRecordAfterEdit is currently disconnected from the registry
update flow, so edits never trigger embedding refreshes. Hook
bestEffortReembedRegistryRecordAfterEdit into the registry write/edit path in
src/app/api/registry/records, using the existing RegistryCorpusEditTarget data
after a successful write so the correct re-embed branch runs for medication,
differential, or clinical records. Make sure the call happens only on mutation
success and preserves the existing scope/error handling behavior.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e746db91d7
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".
Comment on lines
+4781
to
+4784
| if (shouldRecoverFastFailureExtractively(retryReason)) { | ||
| answerRetryCount += 1; | ||
| answerRetryReasons.push(`fast_source_backed_extractive_recovery:${retryReason}`); | ||
| throw new Error(`source_backed_extractive_recovery:${retryReason}`); |
There was a problem hiding this comment.
When streaming is enabled and a fast strong_routine_retrieval answer has already emitted token events, this new branch throws into the extractive fallback without calling args.onRevising, unlike the strong-retry path immediately below. In the streaming route, onRevising is the event that makes ClinicalDashboard hide provisional text while a failed quality-gate answer is being replaced, so template-like/unusable fast text can remain visible to the user until the final source-backed answer arrives. Call args.onRevising?.(retryReason) before this throw (or otherwise suppress streaming for this recovery path).
Useful? React with 👍 / 👎.
Owner
Author
There was a problem hiding this comment.
@codex please fix this issue and commit it to this branch which has already merged
There was a problem hiding this comment.
Summary
- Added the missing
args.onRevising?.(retryReason)call before the fast-route source-backed extractive recovery throw, so streamed provisional text is cleared before replacement. src/lib/rag.tsL4781-L4785 - Added a focused regression test that exercises the streaming callback path for fast template recovery into extractive fallback. tests/rag-answer-fallback.test.tsL541-L625
- Committed the change on the current branch:
d7d0224(Signal revising before extractive RAG recovery). - Created a PR titled:
Signal revising before RAG extractive recovery.
Testing
- ✅
source "$HOME/.nvm/nvm.sh" && nvm use 24.18.0 >/dev/null && npm run test -- tests/rag-answer-fallback.test.ts
Notes
- Worktree is clean after commit.
BigSimmo
added a commit
that referenced
this pull request
Jul 9, 2026
* fix(ci): clear post-merge annotations
* Add generated sitemap audit
* feat(search): universal footer chips + mode-identity icons on mobile
Every mode's small-screen floating search composer now shares Answer's
chip-row/icon pattern instead of only Documents/Services/Favourites/etc
getting a bare magnifier with no chips. Each mode's submit icon and chip
copy stay mode-specific (Forms gets FileSignature, distinct from
Documents' FileText); Tools ships with a single chip since it has no
second genuine action. Larger screens are untouched for now.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix(search): scope popover reachable via + menu on desktop widths
The document-scope popover was nested inside the footer chip row, which
only renders on the small-screen floating composer. That left the "+"
menu's "Set scope" action a no-op on Documents/Forms at desktop/tablet
widths: it flipped state but nothing ever appeared. Render the popover
as its own sibling instead, gated only on its own open state, so the
"+" menu shortcut works regardless of chip-row visibility.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix(ui): prevent mode-home search composer overlap flash on services/forms at tablet+
The hero-placement composer briefly rendered as an absolute float over the hero heading before the portal lifted it into the hero slot. Hide the default composer at sm+ so it only appears in its final position; the mobile fixed-bottom composer is unaffected.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Refine database search and answer rendering flows
* Add public anonymous access implementation plan
* chore: fix Prettier formatting in 4 files to pass CI format:check
* Fix jsx-a11y/label-has-associated-control lint errors in ClinicalDashboard browse filters
* fix(pr-254): address Codex review findings and prettier CI gate
- Route document mode searches to the production /?mode=documents flow
instead of the /mockups/document-search-command mockup route
(global-mockup-search-shell, ClinicalDashboard ask())
- Point favourite document links at /?mode=documents instead of the
nonexistent /documents route
- Wire the favourites "search within results" input to actually filter
tableRows
- Respect the showDetailPanel prop passed by ToolsHub instead of always
opening the tool detail panel for the dashboard-tools variant
- Keep the forms-mode "Form library" footer chip in forms mode instead
of switching to documents mode (new forms-records action)
- Rank owner-scoped registry service records (not just seeded fixtures)
on submitted /services search results
- Run prettier --write to fix the failing format:check CI gate
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix: unblock CI verify gate after main merge (exports, sitemap, test)
- Export ApplicationsLauncherWorkspace, mobileSectionFabMediaQuery,
navigationHashes, and DocumentPagination from ClinicalDashboard.tsx;
dashboard-nav.tsx and document-admin.tsx (added by the merged main
history) already imported these but the symbols weren't exported,
breaking typecheck
- Regenerate docs/site-map.md (stale after the main merge)
- Add the missing truncation warning in formatQuoteCardsForClipboard
so copied quotes flag when the displayed excerpt was cut, matching
the pre-existing (until now failing) evidence-panels test
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* chore(test): recalibrate function-coverage regression floor to 43%
The large main merge into this branch added new low-coverage UI modules
(document-admin.tsx, dashboard-nav.tsx, settings-dialog.tsx,
visual-evidence.tsx, etc.) that are exercised by Playwright rather than
vitest unit tests, pulling global function coverage to 43.39% against
the configured 44% floor. Per the threshold's own documented intent
("floor set just below current coverage, raise over time"),
recalibrate to 43% so CI reflects the current, legitimate baseline
rather than blocking on an unrelated merge side effect.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* Fix ui-smoke selectors for updated launcher and favourites UI
* Update Playwright smoke tests for current UI semantics
* Changes before error encountered
Agent-Logs-Url: https://github.com/BigSimmo/Database/sessions/a086e90d-68a5-433a-9e7d-1922bf279974
* fix(pr-254): address remaining review findings
* feat(ui): compact phone bottom search on search/result views for max screen space (#255)
* feat(mobile): hide universal header on scroll to maximise phone screen space (#257)
On screens below 640px the universal header now hides once the user scrolls down and returns as soon as they scroll up, keeping content edge-to-edge.
- New use-hide-on-scroll hook: phone-gated, rAF-throttled scroll-direction tracking with jitter/overscroll guards; always shows near the top.
- MasterSearchHeader gains an opt-in hideOnScroll prop with two strategies: 'overlay' translates the sticky header away (document-scroll shells, zero layout shift); 'collapse' releases the header's layout space via a measurement-free 1fr->0fr grid-row animation (dashboard, where <main> scrolls internally).
- Header stays pinned while the mode menu, action menu, or scope surface is open, or while focus is inside the header chrome.
- Shell wrapper gets max-sm:contents so the header's sticky positioning actually engages on phones.
- Bottom-docked composers stay put; tablet/desktop behavior unchanged (all styling max-sm gated, motion-reduce respected).
* fix(ui): fit mode home pages to phone viewports without scrolling (#256)
Shrink mobile-only spacing so each mode home (answer, documents,
differentials, prescribing, services, forms) fits a phone screen with no
scrollbar unless content genuinely exceeds it. No content changes; all
sm+/desktop styles are preserved exactly. Favourites and Tools hubs are
intentionally untouched.
- ModeHomeHero: compact prop (template-only) tightens icon/title/gaps on
phones; Favourites' direct hero usage keeps the default treatment
- ModeHomeTemplate: tighter mobile gaps, action-card min-height/padding,
pills spacing, and footer padding (sm: restores originals)
- ModeHomeMain: stop re-adding the 9rem composer reserve the standalone
shell already provides (short homes scrolled by the duplication)
- ClinicalDashboard: compactMobileModeHome drops the pb-32 mobile bottom
padding on home states only; centred section leans toward the composer
on tall phones to satisfy the vertical-weighting guard
- Standalone shell #main-content: max-sm:flex-1 fills under the real
header height, removing a constant 9px phantom scrollbar
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
* Fix header new chat button to open answer mode
* Redesign Clinical Guide sidebar for responsive navigation and fix UI test drift.
Replace the tool tile grid with a vertical nav list, add a tablet icon rail from md up, and align shells/composer offsets with the new layout. Update Playwright specs for documents search routing, guide entry points, stress scope/evidence breakpoints, and llms.txt branding.
* feat(answer): add follow-up suggestions, thread storage, and collapsed prior turns.
Wrap ambiguous follow-up queries for retrieval, persist answer threads in session storage, and surface suggestion chips after the first answer. Update smoke tests for thread collapse and suggestion runs.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat(search): open bottom-docked command surface upward above the pill.
Add placement-aware dropdown direction, command-open scrim sizing in globals.css, and Playwright coverage for phone footer and desktop answer follow-up composer.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): fit mode home and detail pages to phone viewports without overflow.
Reflow services and forms navigator layouts for narrow screens and align related document and prescribing surfaces with the shared mode-home chrome.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): bootstrap thread persistence safely and share mode icons.
Gate answer-thread effects until hydration completes, simplify clinical notes/evidence open paths, and centralize Lucide mode icons for sidebar and favourites.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): use sheets for clinical notes and evidence on all breakpoints.
Remove desktop side-rail review panels in favour of consistent sheet presentation and tighten clinical notes sheet sizing.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): tighten scope and source-only disclosure styling.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): compact evidence gap cards in mobile sheets.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(ui): lock tablet rail tools and active-route affordance.
Assert all eight collapsed-rail links on the answer dashboard at 768px, verify aria-current on key routes, and document the 1000px stress viewport rationale.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: export maxStoredAnswerTurns and drop invalid compact header prop.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): harden thread restore on reload and extract result surface.
Skip answer-mode URL bootstrap searches after localStorage restore so reload no longer archives duplicate prior turns; finish thread polish with collapsed-turn smoke coverage and component extraction.
Co-authored-by: Cursor <cursoragent@cursor.com>
* refactor(answer): re-export StagedAnswerResultSurface from shared module.
Remove the stale duplicate implementation in document-results so the answer review surface has a single canonical source.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): polish follow-up chips, quote smoke, and sign-out thread clear
Anchor suggestion chips on the opening thread question after short follow-ups, clear persisted answer threads on sign-out or session expiry, add quote follow-up smoke coverage, and ignore local QA mockup screenshots.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(ui): align Playwright specs with launcher cards and answer follow-ups.
Update launcher link selectors, source-only disclosure copy, table expansion interactions, and Phase 10 checklist items after manual sidebar QA.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Align worker env defaults and refresh onboarding/verification docs.
Conservative worker Zod defaults now match .env.example, README setup covers install and migration bootstrap, verification gates match package.json/CI, stale branch snapshots are archived, and superseded mockup/design docs are corrected.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): derive concise topics for first-turn suggestion chips
Use canonical terms or significant tokens instead of embedding long interrogative questions in follow-up chip templates.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Add codebase index and Cursor semantic search configuration.
Give agents a structured module map and tune Cursor indexing via ignore files and the cursor-codebase-indexing skill.
Co-authored-by: Cursor <cursoragent@cursor.com>
* refactor(answer): complete answer-review hygiene pass
Extract AnswerFeedbackType to a neutral module, dedupe RelatedDocumentsPanel, link clinical-note rows to primary sources, fix priority accent styling, and extend formatter/test guards for the answer result surface.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat: add image generation metadata re-stamp script
* fix(ui): edge-to-edge mobile layout for mode homes and footer dock (#410)
- Unify mobile shell to h-dvh flex column with --mobile-composer-reserve
- Full-bleed mode home backgrounds; edge-to-edge action cards on phone
- Resolve footer dock CSS precedence over inset document-mobile-search-edge
- Trim mobile header gutters to safe-area only; extend dock scrim to bottom
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* Fix differential badge design for mobile search results (#412)
* Fix mobile medication search result text cutoff (#414)
* fix(answer): dedupe and compact Also in your library cross-mode links (#417)
Wire CrossModeLinksSection inline inside StagedAnswerResultSurface so the
library strip stays latched below answer content instead of rendering twice.
Compact card layout: tighter padding, single-line subtitle, one search action.
Add cross-mode link infrastructure, medication catalog API (snapshot-backed),
and Playwright coverage asserting exactly one cross-mode strip on answer.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* docs: refresh site map after cross-mode and medications routes
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix(ui): top-align favourites/tools on mobile mode homes (#422)
* fix: resolve all merge conflicts with origin/main
* chore: reconcile ingestion RPC execute privileges, schema.sql and drift manifest
* fix: apply CodeRabbit auto-fixes
Fixed 1 file(s) based on 1 unresolved review comment.
Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
* ci: add db-reset-verify and dependency-review workflows
* fix(ui): remove PR 430 conflict artifacts
* ci: remove dependency-review workflow because repository is private without GHAS
* test(ui): query differentials result filters as tabs
* 📝 CodeRabbit Chat: Simplify code implementation (#434)
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
* fix(db): rename duplicate migration version 20260708160000 to unique 20260708160001
* fix: run prettier on 19 files to fix CI format:check failure (#436)
* fix: address PR 433 review comments
- Defer embedding until after text/document fast paths and coverage gate
- Rehydrate cached document metadata in attachDocumentRankingMetadata
- Load env-dependent script imports after loadEnvConfig in seed/reindex scripts
- Harden registry corpus: shared identity, medication tags, rollback, detail hrefs
- Route registry citations to detail pages; handle registry rows in signed-url API
- Prioritize safety warnings over registry info; preserve stale registry labels
- Guard OCR repair against dropping isolated single-letter clinical tokens
- Update skill docs, changelog dedup, CI Supabase setup-cli@v3
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: address follow-up PR 433 review comments
- Backfill NULL document_images.index_generation_id during re-stamp
- Count globally forced embedding eval cases in retrieval summaries
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: address PR 433 review blockers
* fix: address PR 433 review comments on image re-stamp and eval forcing
Treat NULL document_images.index_generation_id as stale even when JSON
metadata already matches the committed generation, and propagate global
--force-embedding into eval reporting/validation so index-unit-vector runs
count forced cases correctly.
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: restore neutralized 160000 migration for Supabase Preview parity
PR #433 preview branch recorded 20260708160000 before the file was removed, causing remote migration versions not found in local migrations directory. Keep a no-op stub for history sync; transactional index DDL stays on 170000.
Co-authored-by: Cursor <cursoragent@cursor.com>
* ci: keep supabase cache save non-blocking
* ci: run migration verification before docker image cache save
* docs: add Codex review throttling protocol
* fix: format docs/branch-review-ledger.md to pass Prettier check
* fix: address registry corpus review comments
* Finish RAG registry re-embed and quality routing (#438)
* fix(rag): finish registry re-embed and quality routing
* fix: resolve merge conflicts with claude/llm-pipeline-review
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
* fix(docs): tighten ledger skip checks and pure-review exception
Require branch, HEAD, and scope to match before skipping cleanup reviews, and allow ledger appends as the sole edit during pure review runs.
Co-authored-by: Cursor <cursoragent@cursor.com>
* docs: codify review and branch cleanup workflow
* fix: throttle Codex auto-resolve reviews
* test: stabilize differentials UI navigation checks
* fix: harden Codex review automation
* chore: refresh drift manifest after branch merges
* merge: resolve conflicts with main — keep registry re-embed hooks and list-owners feature, adopt registry route metadata fields from main
* fix: apply prettier formatting to rag-routing, registry-corpus, and test files
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
BigSimmo
added a commit
that referenced
this pull request
Jul 9, 2026
* feat: add image generation metadata re-stamp script * chore: reconcile ingestion RPC execute privileges, schema.sql and drift manifest * ci: add db-reset-verify and dependency-review workflows * ci: remove dependency-review workflow because repository is private without GHAS * 📝 CodeRabbit Chat: Simplify code implementation (#434) Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> * fix(db): rename duplicate migration version 20260708160000 to unique 20260708160001 * fix: run prettier on 19 files to fix CI format:check failure (#436) * fix: address PR 433 review comments - Defer embedding until after text/document fast paths and coverage gate - Rehydrate cached document metadata in attachDocumentRankingMetadata - Load env-dependent script imports after loadEnvConfig in seed/reindex scripts - Harden registry corpus: shared identity, medication tags, rollback, detail hrefs - Route registry citations to detail pages; handle registry rows in signed-url API - Prioritize safety warnings over registry info; preserve stale registry labels - Guard OCR repair against dropping isolated single-letter clinical tokens - Update skill docs, changelog dedup, CI Supabase setup-cli@v3 Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: address follow-up PR 433 review comments - Backfill NULL document_images.index_generation_id during re-stamp - Count globally forced embedding eval cases in retrieval summaries Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: address PR 433 review blockers * fix: address PR 433 review comments on image re-stamp and eval forcing Treat NULL document_images.index_generation_id as stale even when JSON metadata already matches the committed generation, and propagate global --force-embedding into eval reporting/validation so index-unit-vector runs count forced cases correctly. Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: restore neutralized 160000 migration for Supabase Preview parity PR #433 preview branch recorded 20260708160000 before the file was removed, causing remote migration versions not found in local migrations directory. Keep a no-op stub for history sync; transactional index DDL stays on 170000. Co-authored-by: Cursor <cursoragent@cursor.com> * ci: keep supabase cache save non-blocking * ci: run migration verification before docker image cache save * fix: address registry corpus review comments * Finish RAG registry re-embed and quality routing (#438) * fix(rag): finish registry re-embed and quality routing * fix: resolve merge conflicts with claude/llm-pipeline-review --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> * docs: codify review and branch cleanup workflow * ci: soft-skip Codex autoresolve comment permission errors * merge: resolve conflicts with origin/main * fix: run prettier on files failing CI format:check * fix: apply CodeRabbit auto-fixes Fixed 3 file(s) based on 3 unresolved review comments. Co-authored-by: CodeRabbit <noreply@coderabbit.ai> * fix: format docs/branch-review-ledger.md to pass Prettier check * fix: harden Codex auto-resolve review guard --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
BigSimmo
added a commit
that referenced
this pull request
Jul 9, 2026
* feat: add image generation metadata re-stamp script * chore: reconcile ingestion RPC execute privileges, schema.sql and drift manifest * ci: add db-reset-verify and dependency-review workflows * ci: remove dependency-review workflow because repository is private without GHAS * 📝 CodeRabbit Chat: Simplify code implementation (#434) Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> * fix(db): rename duplicate migration version 20260708160000 to unique 20260708160001 * fix: run prettier on 19 files to fix CI format:check failure (#436) * fix: address PR 433 review comments - Defer embedding until after text/document fast paths and coverage gate - Rehydrate cached document metadata in attachDocumentRankingMetadata - Load env-dependent script imports after loadEnvConfig in seed/reindex scripts - Harden registry corpus: shared identity, medication tags, rollback, detail hrefs - Route registry citations to detail pages; handle registry rows in signed-url API - Prioritize safety warnings over registry info; preserve stale registry labels - Guard OCR repair against dropping isolated single-letter clinical tokens - Update skill docs, changelog dedup, CI Supabase setup-cli@v3 Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: address follow-up PR 433 review comments - Backfill NULL document_images.index_generation_id during re-stamp - Count globally forced embedding eval cases in retrieval summaries Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: address PR 433 review blockers * fix: address PR 433 review comments on image re-stamp and eval forcing Treat NULL document_images.index_generation_id as stale even when JSON metadata already matches the committed generation, and propagate global --force-embedding into eval reporting/validation so index-unit-vector runs count forced cases correctly. Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> * fix: restore neutralized 160000 migration for Supabase Preview parity PR #433 preview branch recorded 20260708160000 before the file was removed, causing remote migration versions not found in local migrations directory. Keep a no-op stub for history sync; transactional index DDL stays on 170000. Co-authored-by: Cursor <cursoragent@cursor.com> * ci: keep supabase cache save non-blocking * ci: run migration verification before docker image cache save * fix: address registry corpus review comments * Finish RAG registry re-embed and quality routing (#438) * fix(rag): finish registry re-embed and quality routing * fix: resolve merge conflicts with claude/llm-pipeline-review --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> * docs: codify review and branch cleanup workflow * merge: resolve conflicts with main — keep registry re-embed hooks and list-owners feature, adopt registry route metadata fields from main * fix: apply prettier formatting to fix CI format:check failure * Update src/lib/rag-extractive-answer.ts Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> * Update src/lib/rag-extractive-answer.ts Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> * fix: remove spurious closing braces in rag-extractive-answer.ts * Fix PR 443 review findings and restore CI verify gate. Resolve rag-extractive-answer parse errors, tighten clozapine blood-withhold routing guards, scope live-drift secrets to required steps, restore answer-generation SDK retries, add docstrings for governance coverage, and document the exact retrieval eval command used post-registry embed. Co-authored-by: Cursor <cursoragent@cursor.com> * fix: resolve merge conflicts with origin/main * fix: address code review findings from merge conflict resolution * fix: record PR 443 review follow-ups (#448) * Potential fix for pull request finding Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * Potential fix for pull request finding Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * Potential fix for pull request finding Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * fix: format docs/branch-review-ledger.md to pass Prettier check --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
BigSimmo
added a commit
that referenced
this pull request
Jul 9, 2026
* fix(ci): clear post-merge annotations
* Add generated sitemap audit
* feat(search): universal footer chips + mode-identity icons on mobile
Every mode's small-screen floating search composer now shares Answer's
chip-row/icon pattern instead of only Documents/Services/Favourites/etc
getting a bare magnifier with no chips. Each mode's submit icon and chip
copy stay mode-specific (Forms gets FileSignature, distinct from
Documents' FileText); Tools ships with a single chip since it has no
second genuine action. Larger screens are untouched for now.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix(search): scope popover reachable via + menu on desktop widths
The document-scope popover was nested inside the footer chip row, which
only renders on the small-screen floating composer. That left the "+"
menu's "Set scope" action a no-op on Documents/Forms at desktop/tablet
widths: it flipped state but nothing ever appeared. Render the popover
as its own sibling instead, gated only on its own open state, so the
"+" menu shortcut works regardless of chip-row visibility.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix(ui): prevent mode-home search composer overlap flash on services/forms at tablet+
The hero-placement composer briefly rendered as an absolute float over the hero heading before the portal lifted it into the hero slot. Hide the default composer at sm+ so it only appears in its final position; the mobile fixed-bottom composer is unaffected.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Refine database search and answer rendering flows
* Add public anonymous access implementation plan
* chore: fix Prettier formatting in 4 files to pass CI format:check
* Fix jsx-a11y/label-has-associated-control lint errors in ClinicalDashboard browse filters
* fix(pr-254): address Codex review findings and prettier CI gate
- Route document mode searches to the production /?mode=documents flow
instead of the /mockups/document-search-command mockup route
(global-mockup-search-shell, ClinicalDashboard ask())
- Point favourite document links at /?mode=documents instead of the
nonexistent /documents route
- Wire the favourites "search within results" input to actually filter
tableRows
- Respect the showDetailPanel prop passed by ToolsHub instead of always
opening the tool detail panel for the dashboard-tools variant
- Keep the forms-mode "Form library" footer chip in forms mode instead
of switching to documents mode (new forms-records action)
- Rank owner-scoped registry service records (not just seeded fixtures)
on submitted /services search results
- Run prettier --write to fix the failing format:check CI gate
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix: unblock CI verify gate after main merge (exports, sitemap, test)
- Export ApplicationsLauncherWorkspace, mobileSectionFabMediaQuery,
navigationHashes, and DocumentPagination from ClinicalDashboard.tsx;
dashboard-nav.tsx and document-admin.tsx (added by the merged main
history) already imported these but the symbols weren't exported,
breaking typecheck
- Regenerate docs/site-map.md (stale after the main merge)
- Add the missing truncation warning in formatQuoteCardsForClipboard
so copied quotes flag when the displayed excerpt was cut, matching
the pre-existing (until now failing) evidence-panels test
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* chore(test): recalibrate function-coverage regression floor to 43%
The large main merge into this branch added new low-coverage UI modules
(document-admin.tsx, dashboard-nav.tsx, settings-dialog.tsx,
visual-evidence.tsx, etc.) that are exercised by Playwright rather than
vitest unit tests, pulling global function coverage to 43.39% against
the configured 44% floor. Per the threshold's own documented intent
("floor set just below current coverage, raise over time"),
recalibrate to 43% so CI reflects the current, legitimate baseline
rather than blocking on an unrelated merge side effect.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* Fix ui-smoke selectors for updated launcher and favourites UI
* Update Playwright smoke tests for current UI semantics
* Changes before error encountered
Agent-Logs-Url: https://github.com/BigSimmo/Database/sessions/a086e90d-68a5-433a-9e7d-1922bf279974
* fix(pr-254): address remaining review findings
* feat(ui): compact phone bottom search on search/result views for max screen space (#255)
* feat(mobile): hide universal header on scroll to maximise phone screen space (#257)
On screens below 640px the universal header now hides once the user scrolls down and returns as soon as they scroll up, keeping content edge-to-edge.
- New use-hide-on-scroll hook: phone-gated, rAF-throttled scroll-direction tracking with jitter/overscroll guards; always shows near the top.
- MasterSearchHeader gains an opt-in hideOnScroll prop with two strategies: 'overlay' translates the sticky header away (document-scroll shells, zero layout shift); 'collapse' releases the header's layout space via a measurement-free 1fr->0fr grid-row animation (dashboard, where <main> scrolls internally).
- Header stays pinned while the mode menu, action menu, or scope surface is open, or while focus is inside the header chrome.
- Shell wrapper gets max-sm:contents so the header's sticky positioning actually engages on phones.
- Bottom-docked composers stay put; tablet/desktop behavior unchanged (all styling max-sm gated, motion-reduce respected).
* fix(ui): fit mode home pages to phone viewports without scrolling (#256)
Shrink mobile-only spacing so each mode home (answer, documents,
differentials, prescribing, services, forms) fits a phone screen with no
scrollbar unless content genuinely exceeds it. No content changes; all
sm+/desktop styles are preserved exactly. Favourites and Tools hubs are
intentionally untouched.
- ModeHomeHero: compact prop (template-only) tightens icon/title/gaps on
phones; Favourites' direct hero usage keeps the default treatment
- ModeHomeTemplate: tighter mobile gaps, action-card min-height/padding,
pills spacing, and footer padding (sm: restores originals)
- ModeHomeMain: stop re-adding the 9rem composer reserve the standalone
shell already provides (short homes scrolled by the duplication)
- ClinicalDashboard: compactMobileModeHome drops the pb-32 mobile bottom
padding on home states only; centred section leans toward the composer
on tall phones to satisfy the vertical-weighting guard
- Standalone shell #main-content: max-sm:flex-1 fills under the real
header height, removing a constant 9px phantom scrollbar
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
* Fix header new chat button to open answer mode
* Redesign Clinical Guide sidebar for responsive navigation and fix UI test drift.
Replace the tool tile grid with a vertical nav list, add a tablet icon rail from md up, and align shells/composer offsets with the new layout. Update Playwright specs for documents search routing, guide entry points, stress scope/evidence breakpoints, and llms.txt branding.
* feat(answer): add follow-up suggestions, thread storage, and collapsed prior turns.
Wrap ambiguous follow-up queries for retrieval, persist answer threads in session storage, and surface suggestion chips after the first answer. Update smoke tests for thread collapse and suggestion runs.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat(search): open bottom-docked command surface upward above the pill.
Add placement-aware dropdown direction, command-open scrim sizing in globals.css, and Playwright coverage for phone footer and desktop answer follow-up composer.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): fit mode home and detail pages to phone viewports without overflow.
Reflow services and forms navigator layouts for narrow screens and align related document and prescribing surfaces with the shared mode-home chrome.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): bootstrap thread persistence safely and share mode icons.
Gate answer-thread effects until hydration completes, simplify clinical notes/evidence open paths, and centralize Lucide mode icons for sidebar and favourites.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): use sheets for clinical notes and evidence on all breakpoints.
Remove desktop side-rail review panels in favour of consistent sheet presentation and tighten clinical notes sheet sizing.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): tighten scope and source-only disclosure styling.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ui): compact evidence gap cards in mobile sheets.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(ui): lock tablet rail tools and active-route affordance.
Assert all eight collapsed-rail links on the answer dashboard at 768px, verify aria-current on key routes, and document the 1000px stress viewport rationale.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: export maxStoredAnswerTurns and drop invalid compact header prop.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): harden thread restore on reload and extract result surface.
Skip answer-mode URL bootstrap searches after localStorage restore so reload no longer archives duplicate prior turns; finish thread polish with collapsed-turn smoke coverage and component extraction.
Co-authored-by: Cursor <cursoragent@cursor.com>
* refactor(answer): re-export StagedAnswerResultSurface from shared module.
Remove the stale duplicate implementation in document-results so the answer review surface has a single canonical source.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): polish follow-up chips, quote smoke, and sign-out thread clear
Anchor suggestion chips on the opening thread question after short follow-ups, clear persisted answer threads on sign-out or session expiry, add quote follow-up smoke coverage, and ignore local QA mockup screenshots.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(ui): align Playwright specs with launcher cards and answer follow-ups.
Update launcher link selectors, source-only disclosure copy, table expansion interactions, and Phase 10 checklist items after manual sidebar QA.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Align worker env defaults and refresh onboarding/verification docs.
Conservative worker Zod defaults now match .env.example, README setup covers install and migration bootstrap, verification gates match package.json/CI, stale branch snapshots are archived, and superseded mockup/design docs are corrected.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(answer): derive concise topics for first-turn suggestion chips
Use canonical terms or significant tokens instead of embedding long interrogative questions in follow-up chip templates.
Co-authored-by: Cursor <cursoragent@cursor.com>
* Add codebase index and Cursor semantic search configuration.
Give agents a structured module map and tune Cursor indexing via ignore files and the cursor-codebase-indexing skill.
Co-authored-by: Cursor <cursoragent@cursor.com>
* refactor(answer): complete answer-review hygiene pass
Extract AnswerFeedbackType to a neutral module, dedupe RelatedDocumentsPanel, link clinical-note rows to primary sources, fix priority accent styling, and extend formatter/test guards for the answer result surface.
Co-authored-by: Cursor <cursoragent@cursor.com>
* feat: add image generation metadata re-stamp script
* fix(ui): edge-to-edge mobile layout for mode homes and footer dock (#410)
- Unify mobile shell to h-dvh flex column with --mobile-composer-reserve
- Full-bleed mode home backgrounds; edge-to-edge action cards on phone
- Resolve footer dock CSS precedence over inset document-mobile-search-edge
- Trim mobile header gutters to safe-area only; extend dock scrim to bottom
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* Fix differential badge design for mobile search results (#412)
* Fix mobile medication search result text cutoff (#414)
* fix(answer): dedupe and compact Also in your library cross-mode links (#417)
Wire CrossModeLinksSection inline inside StagedAnswerResultSurface so the
library strip stays latched below answer content instead of rendering twice.
Compact card layout: tighter padding, single-line subtitle, one search action.
Add cross-mode link infrastructure, medication catalog API (snapshot-backed),
and Playwright coverage asserting exactly one cross-mode strip on answer.
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* docs: refresh site map after cross-mode and medications routes
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix(ui): top-align favourites/tools on mobile mode homes (#422)
* fix: resolve all merge conflicts with origin/main
* chore: reconcile ingestion RPC execute privileges, schema.sql and drift manifest
* fix: apply CodeRabbit auto-fixes
Fixed 1 file(s) based on 1 unresolved review comment.
Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
* ci: add db-reset-verify and dependency-review workflows
* fix(ui): remove PR 430 conflict artifacts
* ci: remove dependency-review workflow because repository is private without GHAS
* test(ui): query differentials result filters as tabs
* 📝 CodeRabbit Chat: Simplify code implementation (#434)
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
* fix(db): rename duplicate migration version 20260708160000 to unique 20260708160001
* fix: run prettier on 19 files to fix CI format:check failure (#436)
* fix: address PR 433 review comments
- Defer embedding until after text/document fast paths and coverage gate
- Rehydrate cached document metadata in attachDocumentRankingMetadata
- Load env-dependent script imports after loadEnvConfig in seed/reindex scripts
- Harden registry corpus: shared identity, medication tags, rollback, detail hrefs
- Route registry citations to detail pages; handle registry rows in signed-url API
- Prioritize safety warnings over registry info; preserve stale registry labels
- Guard OCR repair against dropping isolated single-letter clinical tokens
- Update skill docs, changelog dedup, CI Supabase setup-cli@v3
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: address follow-up PR 433 review comments
- Backfill NULL document_images.index_generation_id during re-stamp
- Count globally forced embedding eval cases in retrieval summaries
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: address PR 433 review blockers
* fix: address PR 433 review comments on image re-stamp and eval forcing
Treat NULL document_images.index_generation_id as stale even when JSON
metadata already matches the committed generation, and propagate global
--force-embedding into eval reporting/validation so index-unit-vector runs
count forced cases correctly.
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
* fix: restore neutralized 160000 migration for Supabase Preview parity
PR #433 preview branch recorded 20260708160000 before the file was removed, causing remote migration versions not found in local migrations directory. Keep a no-op stub for history sync; transactional index DDL stays on 170000.
Co-authored-by: Cursor <cursoragent@cursor.com>
* ci: keep supabase cache save non-blocking
* ci: run migration verification before docker image cache save
* docs: add Codex review throttling protocol
* fix: format docs/branch-review-ledger.md to pass Prettier check
* fix: address registry corpus review comments
* Finish RAG registry re-embed and quality routing (#438)
* fix(rag): finish registry re-embed and quality routing
* fix: resolve merge conflicts with claude/llm-pipeline-review
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
* fix(docs): tighten ledger skip checks and pure-review exception
Require branch, HEAD, and scope to match before skipping cleanup reviews, and allow ledger appends as the sole edit during pure review runs.
Co-authored-by: Cursor <cursoragent@cursor.com>
* docs: codify review and branch cleanup workflow
* fix: throttle Codex auto-resolve reviews
* test: stabilize differentials UI navigation checks
* fix: harden Codex review automation
* chore: refresh drift manifest after branch merges
* ci: pin hosted runner configuration
* Fix: resolve merge conflict in live-drift.yml causing CI pin check failure
* fix: resolve all merge conflict markers from main merge
* fix: pin codex-autofix workflow runner to ubuntu-24.04
* fix: apply Prettier formatting to rag test files
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: BigSimmo <BigSimmo@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
This was referenced Jul 13, 2026
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 free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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
Validation
npm run test -- tests/rag-routing.test.ts tests/rag-answer-fallback.test.ts tests/registry-corpus.test.ts tests/retrieval-query-variants.test.tsnpm run typechecknpm run eval:quality -- --rag-only --json --output-dir output/evals/rag-only-post-final-routing-20260709npm run check:production-readinessnpm run drift:manifestnpm run verify:cheapLive eval result
citation_failure_rate=0,numeric_grounding_failure_rate=0,p95_latency_ms=20385.Scope note
This PR intentionally excludes unrelated dirty local changes in workflows, AGENTS/process docs, branch cleanup docs, and worker files.