perf: trim the answer payload to what the client renders (round 2, phase 4)

src/app/api/answer/route.ts

@@ -20,6 +20,7 @@ import {
   sourceGovernanceWarnings,
 } from "@/lib/source-governance";
 import { parseJsonBody } from "@/lib/validation/body";
+import { toClientAnswerPayload } from "@/lib/answer-client-payload";
 import { answerServerTimingEntries, buildServerTimingHeader } from "@/lib/server-timing";
 import { createAdminClient } from "@/lib/supabase/admin";
 import { logAnswerDiagnostics } from "@/lib/answer-telemetry";
@@ -166,7 +167,9 @@ export async function POST(request: Request) {
     );
     return NextResponse.json(
       {
-        ...answer,
+        // Boundary trim only — governance warnings and diagnostics above
+        // consumed the full answer (see answer-client-payload.ts).
+        ...toClientAnswerPayload(answer),
         degradedMode: answerDegradedModeSignal(answer),
         scope: { ...scope, queryMode: answerBody.queryMode },
         sourceGovernanceWarnings: warnings,

src/app/api/answer/stream/route.ts

@@ -8,6 +8,7 @@ import {
   type ApiRateLimitResult,
 } from "@/lib/api-rate-limit";
 import { publicAccessContext } from "@/lib/public-api-access";
+import { toClientAnswerPayload } from "@/lib/answer-client-payload";
 import { answerQuestionWithScope, type AnswerProgressEvent } from "@/lib/rag";
 import { classifyRagQuery } from "@/lib/clinical-search";
 import { annotateSearchResults, buildEvidenceRelevance } from "@/lib/evidence-relevance";
@@ -231,7 +232,9 @@ function streamAnswer(
           }
 
           send("final", {
-            ...answer,
+            // Boundary trim only — governance warnings and diagnostics above
+            // consumed the full answer (see answer-client-payload.ts).
+            ...toClientAnswerPayload(answer),
             degradedMode: answerDegradedModeSignal(answer),
             scope: scope ? { ...scope, queryMode: body.queryMode } : undefined,
             sourceGovernanceWarnings: warnings,

src/lib/answer-client-payload.ts

@@ -0,0 +1,42 @@
+import type { RagAnswer, SearchResult } from "@/lib/types";
+
+// Route-boundary trim of the answer payload. The retrieval pipeline carries
+// full chunk text plus server-only context on every source (adjacent_context
+// for generation packing, memory cards, table facts, index-unit matches,
+// document summaries), but the client renders only a snippet
+// (retrieval_synopsis ?? content) plus identity, scoring, governance, and
+// label fields. Trimming at the route boundary — never inside rag.ts — keeps
+// caches, generation inputs, and eval behavior byte-identical while cutting
+// the final SSE/JSON event the user waits on after the prose has streamed.
+//
+// Ordering contract: sourceGovernanceWarnings and logAnswerDiagnostics consume
+// the FULL answer and must run before this trim (both routes do).
+
+// Longest snippet the source cards can usefully show; the render policy falls
+// back to `content` only when `retrieval_synopsis` is absent.
+const clientSourceContentMaxChars = 700;
+
+function truncateAtWordBoundary(text: string, maxChars: number) {
+  if (text.length <= maxChars) return text;
+  const slice = text.slice(0, maxChars);
+  const lastSpace = slice.lastIndexOf(" ");
+  return `${slice.slice(0, lastSpace > maxChars * 0.6 ? lastSpace : maxChars).trimEnd()}…`;
+}
+
+function trimSourceForClient(source: SearchResult): SearchResult {
+  const trimmed: SearchResult = {
+    ...source,
+    content: truncateAtWordBoundary(source.content ?? "", clientSourceContentMaxChars),
+  };
+  delete trimmed.adjacent_context;
+  delete trimmed.memory_cards;
+  delete trimmed.table_facts;
+  delete trimmed.index_unit;
+  delete trimmed.document_summary;
+  return trimmed;
+}
+
+export function toClientAnswerPayload<T extends Pick<RagAnswer, "sources">>(answer: T): T {
+  if (!answer.sources?.length) return answer;
+  return { ...answer, sources: answer.sources.map(trimSourceForClient) };
+}

tests/answer-client-payload.test.ts

@@ -0,0 +1,85 @@
+import { describe, expect, it } from "vitest";
+
+import { toClientAnswerPayload } from "@/lib/answer-client-payload";
+import type { RagAnswer, SearchResult } from "@/lib/types";
+
+function fullSource(overrides: Partial<SearchResult> = {}): SearchResult {
+  return {
+    id: "chunk-1",
+    document_id: "doc-1",
+    title: "Clozapine monitoring guideline",
+    file_name: "clozapine.pdf",
+    page_number: 4,
+    chunk_index: 7,
+    section_heading: "Monitoring",
+    content: "Full blood count weekly for 18 weeks. ".repeat(60),
+    retrieval_synopsis: "FBC weekly for 18 weeks, then monthly.",
+    image_ids: [],
+    similarity: 0.82,
+    source_metadata: { document_status: "current" } as SearchResult["source_metadata"],
+    adjacent_context: "Preceding paragraph context. ".repeat(20),
+    document_summary: "A long document summary. ".repeat(30),
+    memory_cards: [{ id: "m1" } as never],
+    table_facts: [{ id: "t1" } as never],
+    index_unit: { unit_type: "table" } as never,
+    ...overrides,
+  } as SearchResult;
+}
+
+function answerWith(sources: SearchResult[]): Pick<RagAnswer, "sources"> {
+  return { sources };
+}
+
+describe("toClientAnswerPayload", () => {
+  it("drops server-only per-source fields the client never renders", () => {
+    const trimmed = toClientAnswerPayload(answerWith([fullSource()])).sources![0];
+    expect(trimmed.adjacent_context).toBeUndefined();
+    expect(trimmed.memory_cards).toBeUndefined();
+    expect(trimmed.table_facts).toBeUndefined();
+    expect(trimmed.index_unit).toBeUndefined();
+    expect(trimmed.document_summary).toBeUndefined();
+  });
+
+  it("keeps identity, snippet, scoring, and governance fields intact", () => {
+    const trimmed = toClientAnswerPayload(answerWith([fullSource()])).sources![0];
+    expect(trimmed.id).toBe("chunk-1");
+    expect(trimmed.title).toBe("Clozapine monitoring guideline");
+    expect(trimmed.retrieval_synopsis).toBe("FBC weekly for 18 weeks, then monthly.");
+    expect(trimmed.similarity).toBe(0.82);
+    expect(trimmed.source_metadata).toEqual({ document_status: "current" });
+    expect(trimmed.page_number).toBe(4);
+  });
+
+  it("truncates long chunk content at a word boundary", () => {
+    const trimmed = toClientAnswerPayload(answerWith([fullSource()])).sources![0];
+    expect(trimmed.content.length).toBeLessThanOrEqual(701);
+    expect(trimmed.content.endsWith("…")).toBe(true);
+    // Word-boundary cut: no partial trailing token before the ellipsis.
+    expect(trimmed.content.at(-2)).not.toBe(" ");
+  });
+
+  it("leaves short content untouched", () => {
+    const short = fullSource({ content: "Short snippet." });
+    expect(toClientAnswerPayload(answerWith([short])).sources![0].content).toBe("Short snippet.");
+  });
+
+  it("does not mutate the original answer (caches keep the full sources)", () => {
+    const source = fullSource();
+    const answer = answerWith([source]);
+    toClientAnswerPayload(answer);
+    expect(answer.sources![0].adjacent_context).toBeTruthy();
+    expect(answer.sources![0].content.length).toBeGreaterThan(700);
+  });
+
+  it("passes through answers without sources", () => {
+    const empty = answerWith([]);
+    expect(toClientAnswerPayload(empty)).toBe(empty);
+  });
+
+  it("materially shrinks a representative payload", () => {
+    const answer = answerWith(Array.from({ length: 8 }, (_, index) => fullSource({ id: `chunk-${index}` })));
+    const fullBytes = JSON.stringify(answer).length;
+    const trimmedBytes = JSON.stringify(toClientAnswerPayload(answer)).length;
+    expect(trimmedBytes).toBeLessThan(fullBytes * 0.5);
+  });
+});