fix(rag): exempt deterministic document-list/table answers from prose sanitizer

src/lib/rag.ts

@@ -4668,6 +4668,7 @@ function buildTableOrVisualSourceLookupAnswer(args: { query: string; results: Se
   return {
     answer,
     citationChunkIds: [source.id],
+    preformatted: true,
     answerSections: [
       {
         heading: "Source match",
@@ -4700,6 +4701,7 @@ function buildDocumentSupportListAnswer(args: { query: string; results: SearchRe
       : `I found ${names.length} indexed documents that support this query: ${names.slice(0, -1).join("; ")}; and ${names.at(-1)}.`;
   return {
     answer,
+    preformatted: true,
     citationChunkIds: Array.from(
       new Set(
         documents.flatMap((document) =>
@@ -4812,6 +4814,7 @@ function buildExtractiveAnswer(args: {
     sources: args.results,
     modelUsed: null,
     routingMode: "extractive",
+    preformatted: hasExtractedAnswer && Boolean((naturalAnswer as { preformatted?: boolean }).preformatted),
     routingReason: args.routeReason,
     queryClass: args.queryClass,
     latencyTimings: args.timings,
@@ -5237,6 +5240,14 @@ function finalizeRagAnswerQuality(answer: RagAnswer, query: string, queryClass:
 }
 
 function finalizeRagAnswerQualityCore(answer: RagAnswer, query: string, queryClass: RagQueryClass): RagAnswer {
+  // Deterministic, template-built answers (document-support lists, table/visual source
+  // references) are well-formed by construction and carry no free-text clinical claims.
+  // The clinical-prose sanitizer/quality gate below is designed for model prose and would
+  // strip their document names (facility codes like "(NOCC)(AKG)" read as non-prose),
+  // turning a valid answer into garble that then fails the gate. Return them untouched.
+  if (answer.preformatted && answer.grounded) {
+    return answer;
+  }
   const cleanedAnswer = sanitizeAnswerText(answer.answer);
   const gapLikeAnswer =
     /could not find enough clean|no relevant clinical source|no current source|cannot provide a clinical answer|cannot provide a source-backed clinical answer|nearby indexed passages|not strong enough to support a reliable answer|no specific\b.*\bcan be confirmed|do not contain indexed guidance|do not contain (?:specific\s+)?information|do not provide specific|no\b.*\bguidance\b.*\bincluded|defer to other sources/i.test(

src/lib/types.ts

@@ -860,6 +860,11 @@ export type RagAnswer = {
   queryClass?: RagQueryClass;
   queryAnalysis?: ClinicalQueryAnalysis;
   responseMode?: AnswerResponseMode;
+  // True for deterministic, template-built answers (document-support lists, table/visual
+  // source references) assembled from source metadata rather than model prose. These are
+  // well-formed by construction, so the clinical-prose sanitizer/quality gate — which would
+  // strip their document names (facility codes read as non-prose) — must be skipped.
+  preformatted?: boolean;
   latencyTimings?: {
     search_cache_hit?: boolean;
     shared_cache_hit?: boolean;

tests/rag-answer-fallback.test.ts

@@ -1588,6 +1588,94 @@ describe("RAG structured-output fallback", () => {
     expect(answer.answer).not.toMatch(/level checks|renal review|baseline tests/i);
   });
 
+  it("preserves parenthetical facility codes in the document-support list answer", async () => {
+    // Regression: the document-list answer is deterministic and pre-formatted, but the
+    // clinical-prose sanitizer used to strip facility-code suffixes like "(NOCC) (AKG)"
+    // (read as non-prose), mangling a valid answer into garble the quality gate then
+    // refused. The `preformatted` flag now exempts it from that sanitizer.
+    vi.stubEnv("OPENAI_API_KEY", "test-key");
+    vi.stubEnv("RAG_SEARCH_CACHE_TTL_MS", "0");
+    vi.stubEnv("RAG_ANSWER_CACHE_TTL_MS", "0");
+
+    const noccDocument = {
+      id: "nocc-doc",
+      title: "National Outcomes and Casemix Collection (NOCC) (AKG)",
+      file_name: "NOCC.pdf",
+      metadata: {
+        source_title: "National Outcomes and Casemix Collection",
+        publisher: "Local service",
+        jurisdiction: "Australia/WA",
+        document_status: "current",
+        clinical_validation_status: "approved",
+        extraction_quality: "good",
+      },
+      text_rank: 0.31,
+    };
+    const noccChunk = {
+      id: "nocc-doc-chunk-1",
+      document_id: "nocc-doc",
+      page_number: 1,
+      chunk_index: 0,
+      section_heading: "Outcome measures",
+      section_path: ["Outcome measures"],
+      content: "National Outcomes and Casemix Collection guidance for outcome measures completion.",
+      retrieval_synopsis: "NOCC source summary.",
+      image_ids: [],
+      text_rank: 0.42,
+    };
+    const rpc = vi.fn(async (name: string) => {
+      if (name === "match_documents_for_query") return { data: [noccDocument], error: null };
+      if (name === "match_document_lookup_chunks_text") return { data: [noccChunk], error: null };
+      if (name === "match_document_chunks_hybrid") {
+        return {
+          data: [
+            source({
+              id: "nocc-doc-chunk-1",
+              document_id: "nocc-doc",
+              title: "National Outcomes and Casemix Collection (NOCC) (AKG)",
+              file_name: "NOCC.pdf",
+              section_heading: "Outcome measures",
+              content: "National Outcomes and Casemix Collection guidance for outcome measures completion.",
+              similarity: 0.91,
+              hybrid_score: 0.93,
+              text_rank: 0.42,
+            }),
+          ],
+          error: null,
+        };
+      }
+      if (name === "get_related_document_metadata") return { data: [], error: null };
+      return { data: [], error: null };
+    });
+
+    vi.doMock("@/lib/supabase/admin", () => ({
+      createAdminClient: () => ({
+        rpc,
+        from: vi.fn(() => new EmptyQuery()),
+      }),
+    }));
+    vi.doMock("@/lib/openai", () => ({
+      embedTextWithTelemetry: vi.fn(async () => ({ embedding: [0.1, 0.2, 0.3], cacheHit: false })),
+      generateStructuredTextResult: vi.fn(),
+    }));
+
+    const { answerQuestionWithScope } = await import("../src/lib/rag");
+
+    const answer = await answerQuestionWithScope({
+      query: "What documents support outcome measures completion?",
+      ownerId: undefined,
+      logQuery: false,
+      skipCache: true,
+    });
+
+    expect(answer.routingMode).toBe("extractive");
+    expect(answer.grounded).toBe(true);
+    expect(answer.preformatted).toBe(true);
+    // The facility-code suffix survives intact (previously mangled to a dangling "(").
+    expect(answer.answer).toContain("(NOCC) (AKG)");
+    expect(answer.answer).not.toMatch(/\(\s*(?:;|$| Outcome)/);
+  });
+
   it("does not promote lithium dosage headings or continuation fragments as the primary answer", async () => {
     vi.stubEnv("OPENAI_API_KEY", "test-key");
     vi.stubEnv("RAG_SEARCH_CACHE_TTL_MS", "0");