Fix clinical-subject dose retrieval

[BigSimmo]

Summary

• require dose evidence to carry the question's clinical subject before the fast path can be accepted
• treat dose-only questions separately from route questions and recognise plural dose wording
• include the clinical subject in medication-dose retrieval selection so unrelated numeric tables cannot outrank subject-matched guidance
• add focused regression coverage for entity-crossing evidence, route handling, ranking, and selection

Verification

• npx vitest run tests/smart-rag-api.test.ts tests/retrieval-query-variants.test.ts tests/retrieval-selection.test.ts tests/clinical-search.test.ts tests/eval-search.test.ts (122 passed)
• npm run eval:rag:offline (21 files, 265 tests passed)
• npm run verify:cheap (210 files passed, 1 skipped; 1,934 tests passed, 1 skipped)
• live production-corpus case opioid-withdrawal-doses: document recall 1.0, content recall 1.0, nDCG@10 0.9912, irrelevant-source rate 0, required-signal coverage 1.0

Safety

• no dependency or schema changes in this PR
• production ranking weights are unchanged; the new signal is scoped to explicit medication dose/route/frequency queries
• no secret values are printed or committed

[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 ↗︎.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: e55ac631-1d81-430e-959e-cf966aaa196c

📥 Commits

Reviewing files that changed from the base of the PR and between cc6bfc1 and b8c5cb7.

📒 Files selected for processing (4)

• src/lib/clinical-search.ts
• src/lib/rag.ts
• src/lib/retrieval-selection.ts
• tests/retrieval-query-variants.test.ts

---

📝 Walkthrough

Summary by CodeRabbit

• Improvements
  • Improved medication dose and risk search results by matching evidence to the specific medication or clinical subject in the query.
  • Prioritized dose information associated with the requested medication over unrelated numeric dosage content.
  • Refined route and frequency handling so only relevant evidence is required for each question.
  • Reduced the chance of accepting dose or route information from an unrelated medication context.

Walkthrough

Medication dose and risk retrieval now extracts clinical-subject terms, matches them against evidence, adjusts ranking, and requires contextual dose or route support in coverage gating. Tests cover co-located evidence, unrelated medication contexts, and subject-aware result ranking.

Changes

Medication dose context

Layer / File(s)	Summary
Subject context and scoring src/lib/clinical-search.ts	Adds subject-token extraction and query-result context matching, then applies matching boosts or mismatch penalties to clinical ranking.
Retrieval subject selection src/lib/retrieval-selection.ts	Expands dose/route detection, adds conditional clinical_subject signals, and boosts candidates with matching clinical context.
Contextual coverage gate and validation src/lib/rag.ts, tests/retrieval-query-variants.test.ts	Requires contextual dose and route evidence where applicable and tests matching, entity isolation, and ranking behavior.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Query
  participant RetrievalSelection
  participant ClinicalSearch
  participant RAGCoverageGate
  participant Evidence
  Query->>RetrievalSelection: build medication dose retrieval intent
  RetrievalSelection->>ClinicalSearch: match clinical subject against Evidence
  ClinicalSearch-->>RetrievalSelection: return subject signal and score adjustment
  RetrievalSelection->>RAGCoverageGate: evaluate selected evidence
  RAGCoverageGate->>ClinicalSearch: check contextual dose and route match
  ClinicalSearch->>Evidence: inspect extracted evidence text
  Evidence-->>RAGCoverageGate: return contextual match status
  RAGCoverageGate-->>Query: accept or reject evidence

Loading

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/opioid-dose-retrieval-gate

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch codex/opioid-dose-retrieval-gate

---

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b8c5cb785e

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".