fix(insights): hand the briefing its figures in the reader's units

The daily briefing serialised the aggregated feature set as computed:
weight, grip strength and the weight target in kilograms, waist in
centimetres, glucose and the signals of the day in mg/dL (the glucose
signal carried no unit at all), workout distances in kilometres. A
reader on pounds or mmol/L got a briefing in units they do not use.

One step now converts the feature set after all canonical arithmetic,
on both the nightly and the on-demand path: absolute values through the
full transform, differences and slopes through the factor alone, every
converted block labelled, workout distances under unit-neutral keys.
The prompt, the stored fingerprint and the grounding check all read the
converted set, so a unit switch moves the fingerprint and the grounding
check compares the reply against the figures the model was shown. The
weight target sentence, the comparison block and the GLP-1 plateau
context quote weight in the reader's unit; the comparison block also
labelled sleep as hours while carrying minutes.

A unit switch now also re-warms a cached briefing alongside the status
notes and narratives.

Refs #1067

src/app/api/insights/generate/__tests__/route.test.ts

@@ -12,7 +12,7 @@
  *   - 429 from the provider → 429 (passthrough, not 5xx)
  *   - any other status → 422 (generic provider-connection failure)
  */
-import { describe, it, expect, vi, beforeEach } from "vitest";
+import { describe, it, expect, vi, beforeEach, onTestFinished } from "vitest";
 
 // Mocks must be hoisted before importing the route.
 vi.mock("@/lib/api-handler", () => ({
@@ -171,6 +171,7 @@ import {
   FeaturesPayloadTooLargeError,
 } from "@/lib/insights/features";
 import { annotate } from "@/lib/logging/context";
+import { buildUserPrompt } from "@/lib/ai/prompts/insight-system-prompt";
 import { AiUnavailableError } from "@/lib/ai/capabilities/refusal";
 
 beforeEach(() => {
@@ -460,6 +461,44 @@ describe("POST /api/insights/generate — cache write (v1.16.8)", () => {
     expect(args.data.insightsCachedLocale).toBe("en");
   });
 
+  it("serialises the feature set in the reader's units", async () => {
+    makeWorkingProvider();
+    const user = {
+      insightsPrivacyMode: "aggregated",
+      insightsCachedAt: null,
+      insightsCachedText: null,
+      locale: "en",
+      unitPreference: "imperial",
+      glucoseUnit: "mmol/L",
+    };
+    const findUnique = vi.mocked(prisma.user.findUnique);
+    const original = findUnique.getMockImplementation();
+    findUnique.mockResolvedValue(user as never);
+    onTestFinished(() => {
+      findUnique.mockImplementation(original!);
+    });
+    vi.mocked(extractFeatures).mockResolvedValueOnce({
+      weight: { latest: 80, avg7: 80.5 },
+      glucose: { latest: 126, avg30: 99 },
+    } as never);
+
+    const res = await POST(jsonRequest({ force: true }) as never);
+    expect(res.status).toBe(200);
+
+    const featuresJson = vi.mocked(buildUserPrompt).mock.calls[0][0];
+    const features = JSON.parse(featuresJson);
+    expect(features.weight).toMatchObject({
+      unit: "lb",
+      latest: 176.4,
+      avg7: 177.5,
+    });
+    expect(features.glucose).toMatchObject({
+      unit: "mmol/L",
+      latest: 7,
+      avg30: 5.5,
+    });
+  });
+
   it("refuses a stale cache write when AI profile inclusion changes during generation", async () => {
     makeWorkingProvider();
     selfContextText

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

@@ -87,6 +87,8 @@ import { resolveServerLocale } from "@/lib/i18n/server-locale";
 import { normalizeLocale } from "@/lib/insights/status-shared";
 import { hasUsableStatusProvider } from "@/lib/insights/status-provider";
 import { hashInsightSnapshot } from "@/lib/insights/snapshot-hash";
+import { featuresInReaderUnits } from "@/lib/insights/features-units";
+import { resolveUnitPreferences } from "@/lib/measurements/display-transform";
 import { enqueueForceWarm } from "@/lib/jobs/insight-pregenerate-shared";
 // v1.28.25 — the comparison-snapshot builder was a private near-copy of
 // the lib export (drifted only in comments); the route now shares the
@@ -340,6 +342,10 @@ export const POST = apiHandler((request: NextRequest) =>
         // v1.25 — per-user response-timeout (seconds), threaded onto the
         // generation calls so a slow self-hosted backend honours the setting.
         aiResponseTimeoutSeconds: true,
+        // The reader's units; the feature set is converted into them before
+        // it is serialised, hashed and checked for grounding.
+        unitPreference: true,
+        glucoseUnit: true,
       },
     });
     const modeAtRead = dbUser?.insightsPrivacyMode ?? "aggregated";
@@ -610,6 +616,13 @@ export const POST = apiHandler((request: NextRequest) =>
     // BEFORE serialisation so the model never sees the dropped blocks.
     const excludeList = dbUser?.insightsExcludeMetrics ?? [];
     features = applyInsightsExcludeFilter(features, excludeList);
+    // The reader's units, after all canonical arithmetic: the prompt, the
+    // stored hash and the grounding check read the same converted figures.
+    const units = resolveUnitPreferences({
+      unitPreference: dbUser?.unitPreference,
+      glucoseUnit: dbUser?.glucoseUnit,
+    });
+    features = featuresInReaderUnits(features, units);
     // v1.4.36 W3 T4 — drop zero-row blocks so the prompt never carries
     // labelled-empty sections (`"sleep": []`, `"medications": []`,
     // etc.). Prevents the model from narrating "there are no
@@ -651,7 +664,11 @@ export const POST = apiHandler((request: NextRequest) =>
       comparisonSnapshot ?? undefined,
     );
     if (plateauContext) {
-      userPrompt += buildGlp1PlateauPrompt(plateauContext, locale);
+      userPrompt += buildGlp1PlateauPrompt(
+        plateauContext,
+        locale,
+        units.system,
+      );
     }
     // v1.15.20 — fold the user-authored "about me" self-description
     // (Settings → AI) into the briefing as a delimited, user-provided