Enforce dose context before text fast path

[BigSimmo]

Summary

• resolve the P1 review finding from Fix clinical-subject dose retrieval #571 by applying the contextual medication-dose coverage gate before the earliest text fast-path return
• require numeric dose amounts only for explicit dose-amount questions, and route evidence for route questions
• preserve non-dose pathway/monitoring behavior even when the broad query classifier selects medication_dose_risk
• treat recommendation/patient wording as query framing rather than extra clinical subjects
• make fallback tests model the embedding rescue path when weak dose evidence is correctly rejected

Verification

• focused clinical retrieval/fallback suites: 6 files, 151 tests passed
• npm run eval:rag:offline: 21 files, 265 tests passed
• TypeScript passed
• targeted ESLint passed
• diff check passed

Review resolution

Fixes the P1 inline finding at #571 discussion r3569692474 by preventing mixed opioid-context/unrelated-dose evidence from bypassing evaluateEvidenceCoverageGate.

[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]

📝 Walkthrough

Walkthrough

Medication dose query filtering now recognizes more generic terms and detects explicit dose-evidence requests. RAG fast-path decisions use contextual dose, amount, route, and agitation evidence checks, with tests covering accepted and rejected contexts.

Changes

Medication dose evidence gating

Layer / File(s)	Summary
Dose evidence query classification src/lib/clinical-search.ts	Medication dose subject filtering excludes additional generic terms, and a helper detects explicit dose, route, and frequency evidence requests.
Contextual fast-path coverage gate src/lib/rag.ts, tests/retrieval-query-variants.test.ts, tests/rag-answer-fallback.test.ts	Medication dose fast-pathing uses contextual coverage checks, while tests cover matching and missing dose context and provide a concrete embedding mock.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant decideTextFastPath
  participant isMedicationDoseEvidenceQuery
  participant evaluateEvidenceCoverageGate
  decideTextFastPath->>isMedicationDoseEvidenceQuery: classify query
  isMedicationDoseEvidenceQuery-->>decideTextFastPath: explicit evidence request
  decideTextFastPath->>evaluateEvidenceCoverageGate: evaluate contextual dose coverage
  evaluateEvidenceCoverageGate-->>decideTextFastPath: accepted decision or failure reason

Loading

---

Caution

Pre-merge checks failed

Please resolve all errors before merging. Addressing warnings is optional.

• Ignore

❌ Failed checks (2 errors)

Check name	Status	Explanation	Resolution
Verification Claims	❌ Error	PR verification bullets claim passes generically ('TypeScript passed', 'targeted ESLint passed', etc.) without exact commands/checks; only eval:rag:offline is named.	Rewrite each verification item as an exact command/check plus result, e.g. 'Ran npm run typecheck: passed' or 'Not run: reason'.
Api Route Failure Handling	❌ Error	Common explicit dose/frequency queries like “how much/how often” still bypass the new gate because isMedicationDoseEvidenceQuery only matches keyword tokens.	Expand the detector to cover plain-language dose/route/frequency phrasing, or route all medication_dose_risk queries through evaluateEvidenceCoverageGate.

✅ Passed checks (9 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is concise and accurately captures the main change: enforcing dose context before the text fast path.
Description check	✅ Passed	Summary, verification, and review-resolution are present and specific to the change.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Generated And Sensitive Files	✅ Passed	The PR only changes workflows, package.json, and regenerated package-lock.json; no real secrets, .env files, build artifacts, or local dumps are present.
Risky Git Or Deployment Actions	✅ Passed	PASS: The changed workflows/package scripts contain no force-push/reset/clean/delete/rebase guidance or unguarded deployment steps; eval-canary explicitly requires a green workflow_dispatch check.
Supabase Project And Schema Safety	✅ Passed	PR only changes clinical-search/RAG logic and tests; no Supabase project refs, env examples, migrations, schema, RLS, or policy files were touched.
Runtime And Package Manager Integrity	✅ Passed	package.json/package-lock still pin npm@11.17.0 and Node 24.x, .npmrc keeps engine-strict=true, and workflows still use npm ci.

✨ 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/dose-fast-path-context

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch codex/dose-fast-path-context

---

_{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: 26c93547d8

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".

[coderabbitai]

Actionable comments posted: 2

🤖 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/clinical-search.ts`:
- Around line 1147-1152: Expand and make the shared
isMedicationDoseEvidenceQuery detector intent-aware, covering explicit amount,
frequency, and administration wording such as “how much,” “how often,”
“micrograms,” and oral/intramuscular administration. In
src/lib/clinical-search.ts lines 1147-1152 update the shared predicate; in
src/lib/rag.ts lines 2612-2615 ensure every matching explicit query enters the
coverage gate; and in src/lib/rag.ts lines 2929-2931 reuse the shared
amount/route predicates so gate requirements remain consistent.

In `@src/lib/rag.ts`:
- Around line 2953-2962: The failure-reason chain in the acceptance result must
only report missing dose amount evidence when the query requests an amount.
Update the `!hasDoseAmount` branch to also require `asksDoseAmount`, preserving
route-only queries so they can correctly return `missing_route_evidence`.

🪄 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: 1121ed1f-b25c-4ec5-bf6a-a6646fd962de

📥 Commits

Reviewing files that changed from the base of the PR and between 343531b and 26c9354.

📒 Files selected for processing (4)

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