src/lib/clinical-search.ts (1)

750-761: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Medication dose unit/frequency vocabularies are duplicated across clinical-search.ts and rag.ts. The amount-unit spelling list (mg/mcg/micrograms/milligrams/ug/µg) and the frequency-token list (once/twice/daily/…/qid/every-N-hours) are each maintained independently in two places; this very PR had to edit both in lockstep to add the same new spellings, which is the classic sign of future drift.

• src/lib/clinical-search.ts#L750-L761: canonical site — export the unit pattern (and ideally the frequency pattern near Line 1190) as shared constants.
• src/lib/clinical-search.ts#L1186-L1191: reuse the same exported unit/frequency constants instead of re-listing the spellings.
• src/lib/rag.ts#L2764-L2766: import the shared unit constant from clinical-search.ts instead of re-declaring the identical regex.
• src/lib/rag.ts#L2775-L2781: import the shared frequency constant from clinical-search.ts instead of re-declaring an overlapping regex.

🤖 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/clinical-search.ts` around lines 750 - 761, Deduplicate medication
vocabulary regexes by exporting shared unit and frequency pattern constants from
clinical-search.ts, anchored at hasDoseEvidenceSupport and the frequency logic
around lines 1186-1191. Update clinical-search.ts to reuse those constants, and
in src/lib/rag.ts lines 2764-2766 and 2775-2781 import and reuse the shared unit
and frequency constants instead of redeclaring the patterns; preserve the
existing matching behavior.

src/lib/rag.ts (1)

2936-2991: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Solid, test-verified co-location logic — consider extracting the repeated contextual-check pattern.

Manually traced this cascade against every provided medication_dose_risk test case (split-evidence rejection, route-only rejection, entity-crossing rejection, dose+route acceptance, microgram-only acceptance) and every reason matches. The five top.some((result) => hasDoseEvidenceSupport(result) && <check> && medicationDoseQueryContext(query, result).matched) blocks differ only by the injected check function, which is a good candidate for a small helper to reduce duplication and ease future attribute additions (e.g., a future "duration" attribute).

♻️ Extract a contextual-evidence helper

+const hasContextualEvidence = (
+  top: SearchResult[],
+  query: string,
+  check: (result: SearchResult) => boolean = () => true,
+) =>
+  top.some(
+    (result) =>
+      hasDoseEvidenceSupport(result) && check(result) && medicationDoseQueryContext(query, result).matched,
+  );
+
-    const hasContextualDoseEvidence = top.some(
-      (result) => hasDoseEvidenceSupport(result) && medicationDoseQueryContext(query, result).matched,
-    );
-    const hasContextualDoseAmount = top.some(
-      (result) =>
-        hasDoseEvidenceSupport(result) &&
-        hasDoseAmountEvidenceForGate(result) &&
-        medicationDoseQueryContext(query, result).matched,
-    );
+    const hasContextualDoseEvidence = hasContextualEvidence(top, query);
+    const hasContextualDoseAmount = hasContextualEvidence(top, query, hasDoseAmountEvidenceForGate);

🤖 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.ts` around lines 2936 - 2991, Extract the repeated contextual
`top.some` checks in the `medication_dose_risk` branch into a small helper that
accepts the result collection and attribute-specific evidence predicate, while
consistently requiring `hasDoseEvidenceSupport` and a matching
`medicationDoseQueryContext`. Replace `hasContextualDoseEvidence`,
`hasContextualDoseAmount`, `hasContextualRoute`, `hasContextualFrequency`, and
the corresponding co-location checks in this branch with the helper, preserving
all existing acceptance and reason behavior.