merge: wave/v1220-grounding into release/v1.21.2

src/app/api/insights/chat/route.ts

@@ -535,6 +535,15 @@ Reply now as the assistant, in ${locale === "de" ? "German" : "English"}.`;
   // v1.21.0 (P6) — the present tool-result payloads this turn, for the post-hoc
   // prose number-verifier. Empty on the no-tools path.
   let toolResultPayloads: unknown[] = [];
+  // v1.21.2 (A8) — no-tools/local-provider parity for the prose number-verifier.
+  // The tool path grades prose against the figures the tools returned; the
+  // no-tools path has no tools, so the authoritative set is the SNAPSHOT the
+  // model was actually shown this turn — `snapshot.sections`, the structured
+  // record `snapshotJson` is serialised from, which already carries the
+  // correlations-snapshot block. Populated only when the full figures were
+  // delivered this turn (`includeFullSnapshot`); on a cheap follow-up the block
+  // was not re-sent, so there is no fresh authoritative set to grade against.
+  let noToolsSnapshotPayloads: unknown[] = [];
   let totalTokensSpent: number;
   // v1.21.0 (F3) — cached-input tokens to subtract at reconcile (prompt-cached
   // input the user did not re-pay for must not be billed to the daily meter).
@@ -614,6 +623,14 @@ Reply now as the assistant, in ${locale === "de" ? "German" : "English"}. Fetch
       workingProviderType = fallback.workingProvider.providerType;
       totalTokensSpent = result.tokensUsed ?? 0;
       cachedTokensSpent = result.cachedInputTokens ?? 0;
+      // v1.21.2 (A8) — the no-tools path showed the model the full SNAPSHOT only
+      // when `includeFullSnapshot` was set; otherwise it shipped the
+      // grounded-elsewhere pointer with no fresh figures, so there is nothing to
+      // grade. When figures WERE delivered, the authoritative set is the
+      // structured snapshot record (incl. the correlations block).
+      if (includeFullSnapshot) {
+        noToolsSnapshotPayloads = [snapshot.sections];
+      }
     }
   } catch (err) {
     // The provider chain failed outright — no tokens were billed, so refund
@@ -737,18 +754,27 @@ Reply now as the assistant, in ${locale === "de" ? "German" : "English"}. Fetch
     });
   }
 
-  // v1.21.0 (P6 / C2-5) — post-hoc numeric verifier on the Coach prose. On the
-  // tool path, cross-check every number the model cited against the figures the
-  // tools actually returned this turn; an unmatched number (transcription /
-  // paraphrase drift) is soft-stripped to "[unverified]" and annotated. Cheap,
-  // non-blocking, and a no-op when no tool returned figures (a qualitative turn
-  // or the no-tools path) — the prompt-level grounding rule remains the
-  // backstop there, exactly like the briefing's "no signals → skip". A blocked
-  // turn already carries canned fallback prose, so skip it.
-  if (!outbound.block && toolResultPayloads.length > 0) {
+  // v1.21.0 (P6 / C2-5) — post-hoc numeric verifier on the Coach prose. Cross-
+  // check every number the model cited against this turn's authoritative figure
+  // set; an unmatched number (transcription / paraphrase drift) is soft-stripped
+  // to "[unverified]" and annotated. Cheap, non-blocking, and a no-op when there
+  // is no authoritative set — the prompt-level grounding rule remains the
+  // backstop, exactly like the briefing's "no signals → skip". A blocked turn
+  // already carries canned fallback prose, so skip it.
+  //
+  // v1.21.2 (A8) — the authoritative set is the figures the tools returned on the
+  // tool path, and the SNAPSHOT the model was shown on the no-tools/local path
+  // (`snapshot.sections`, which already carries the correlations-snapshot block).
+  // Exactly one is populated per turn; the grading, tolerance, and exemptions are
+  // identical, so a number the model invents is flagged the same way on both.
+  const authoritativePayloads =
+    toolResultPayloads.length > 0
+      ? toolResultPayloads
+      : noToolsSnapshotPayloads;
+  if (!outbound.block && authoritativePayloads.length > 0) {
     const unverified = findUnverifiedCoachNumbers(
       replyText,
-      toolResultPayloads,
+      authoritativePayloads,
     );
     if (unverified.length > 0) {
       const { prose: corrected, stripped } = stripUnverifiedNumbers(

src/lib/ai/coach/__tests__/coach-prose-grounding-no-tools.test.ts

@@ -0,0 +1,71 @@
+/**
+ * v1.21.2 (A8) — no-tools/local-provider parity for the prose number-verifier.
+ *
+ * On the no-tools path the Coach has no tools, so the authoritative figure set
+ * is the SNAPSHOT the model was shown — the structured `snapshot.sections`
+ * record `snapshotJson` is serialised from, which already carries the
+ * correlations-snapshot block. This proves the SAME verifier grades prose
+ * against that record: a number present in the snapshot (including a
+ * correlations-block leaf) is not flagged; a number absent from it is flagged
+ * exactly as on the tool path.
+ */
+import { describe, expect, it } from "vitest";
+
+import { findUnverifiedCoachNumbers } from "@/lib/ai/coach/coach-prose-grounding";
+
+/**
+ * A representative `snapshot.sections` record: a derived block plus the
+ * correlations-snapshot block (a `CorrelationsSnapshotBlock`) the no-tools path
+ * folds in. The driver's numeric leaves (`r`, `n`, `pairsTested`, `windowDays`)
+ * are authoritative figures the model may legitimately cite.
+ */
+const snapshotSections: Record<string, unknown> = {
+  bloodPressure: {
+    aggregate: { avgSys30: 128, avgDia30: 82 },
+  },
+  correlations: {
+    drivers: [
+      {
+        behaviour: "sleep duration",
+        outcome: "resting heart rate",
+        direction: "lower",
+        lagDays: 1,
+        n: 24,
+        r: 0.62,
+        note: "More sleep tracked with a lower next-day resting heart rate.",
+      },
+    ],
+    pairsTested: 18,
+    windowDays: 90,
+  },
+};
+
+// The route passes the structured record as a single authoritative payload.
+const noToolsPayloads: unknown[] = [snapshotSections];
+
+describe("no-tools verifier parity", () => {
+  it("does not flag a number that traces to a snapshot leaf", () => {
+    const prose = "Your systolic averaged 128 and diastolic 82 this month.";
+    expect(findUnverifiedCoachNumbers(prose, noToolsPayloads)).toEqual([]);
+  });
+
+  it("does not flag a number that traces to the correlations-snapshot block", () => {
+    // r = 0.62 is a leaf of the correlations block folded into the snapshot.
+    const prose = "Sleep and resting heart rate moved together (r = 0.62).";
+    expect(findUnverifiedCoachNumbers(prose, noToolsPayloads)).toEqual([]);
+  });
+
+  it("flags a number the snapshot never carried", () => {
+    const prose = "Your systolic averaged about 138 recently.";
+    const findings = findUnverifiedCoachNumbers(prose, noToolsPayloads);
+    expect(findings).toHaveLength(1);
+    expect(findings[0].value).toBe(138);
+  });
+
+  it("no-ops when the snapshot was not delivered this turn", () => {
+    // Cheap follow-up: the route leaves the no-tools payload set empty, so even
+    // an invented figure is not graded — the prompt-level rule is the backstop.
+    const prose = "Your systolic averaged 138 recently.";
+    expect(findUnverifiedCoachNumbers(prose, [])).toEqual([]);
+  });
+});