Take the writes off three read routes

A GET must not have side effects — api-handler.ts says so next to
READ_HTTP_METHODS, where an MCP-audience token is admitted on the safe
methods precisely because they are assumed to write nothing. Three routes
broke that: the sync handshake bumped User.lastSyncedAt, the achievements
read inserted unlock rows, and the cycle calendar fired a discarded promise
at the forecast cache so a failed write left no trace anywhere.

Each write moves to the place that causes it.

The sync checkpoint is now stamped by the batch ingest endpoints for
measurements, mood and intakes. The handshake reports it. That also fixes
the watermark: the checkpoint used to advance past a window the client had
not drained, and a client trusting it would skip that window. It can only
trail a real sync now.

The unlock rows move to a nightly achievement-unlock-sweep. The badge grid
never needed them — unlocked comes from the live metrics and the completion
date is derived — so the payload is unchanged; the rows pin that date and
carry it into the backup, and the sweep reaches accounts that never open
the page.

The cycle forecast moves to a nightly cycle-prediction-refresh. The
CyclePrediction row is not a cache of a read: it is the index the reminder
cron scans, so which reminders an account received depended on whether
somebody had opened the calendar. persistPredictionCache no longer
swallows its errors either — the job reports them.

src/app/api/cycle/calendar/__tests__/verdict-round-trip.test.ts

@@ -22,6 +22,12 @@ vi.mock("@/lib/db", () => ({
     menstrualCycle: { findMany: vi.fn() },
     cycleDayLog: { findMany: vi.fn() },
     measurement: { findMany: vi.fn() },
+    // Present so a re-introduced forecast persist is VISIBLE here rather
+    // than blowing up on an undefined model. `persistPredictionCache` calls
+    // `findUnique` synchronously before its first await, so even the old
+    // fire-and-forget `void persistPredictionCache(...)` lands on this mock
+    // before the assertion runs.
+    cyclePrediction: { findUnique: vi.fn(), upsert: vi.fn() },
     auditLog: { create: vi.fn() },
   },
 }));
@@ -34,10 +40,6 @@ vi.mock("@/lib/rate-limit", () => ({
   rateLimitHeaders: () => ({}),
 }));
 vi.mock("@/lib/cycle/gate", () => ({ requireCycleEnabled: vi.fn() }));
-// The cache persist is fire-and-forget and would reach the database.
-vi.mock("@/lib/cycle/prediction-cache", () => ({
-  persistPredictionCache: vi.fn().mockResolvedValue(undefined),
-}));
 vi.mock("@/lib/logging/transports", () => ({ emitIfSampled: vi.fn() }));
 vi.mock("@/lib/db-compat", () => ({
   ensureDbCompatibility: vi.fn().mockResolvedValue(undefined),
@@ -142,6 +144,22 @@ afterEach(() => {
   vi.useRealTimers();
 });
 
+describe("GET /api/cycle/calendar is a read", () => {
+  it("persists no forecast — the reminder cron's row is the job's, not this route's", async () => {
+    vi.useFakeTimers();
+    vi.setSystemTime(new Date("2026-06-10T12:00:00Z"));
+
+    const { status, body } = await callCalendar();
+    expect(status).toBe(200);
+    // There IS a forecast to persist, so the absence of a write is a
+    // decision rather than an empty case.
+    expect((body.data as Record<string, unknown>).prediction).not.toBeNull();
+
+    expect(prisma.cyclePrediction.findUnique).not.toHaveBeenCalled();
+    expect(prisma.cyclePrediction.upsert).not.toHaveBeenCalled();
+  });
+});
+
 describe("GET /api/cycle/calendar carries the resolved verdict", () => {
   it("puts the whole verdict on the wire", async () => {
     // 2026-06-10 12:00 UTC is already 2026-06-11 for a user at UTC+14.

src/app/api/cycle/calendar/route.ts

@@ -3,9 +3,9 @@
  * (ios-contract §2.D).
  *
  * Calls the pure engine (`predictCycle` + per-day phase) to build
- * `{ profile, prediction, days }`. The synchronous engine call is cheap
- * (pure stats); the heavy part is the cache persist, which is debounced
- * stale-while-revalidate and fire-and-forget so it never blocks the GET.
+ * `{ profile, prediction, days }`. The engine call is cheap (pure stats) and
+ * the route writes nothing — the `CyclePrediction` row the reminder cron
+ * reads is owned by the `cycle-prediction-refresh` job.
  * Fertile-window fields are goal-gated (surfaced for TRYING_TO_CONCEIVE and
  * AVOID_PREGNANCY — the latter with the stronger "not a contraceptive method"
  * disclaimer; suppressed for GENERAL_HEALTH / PERIMENOPAUSE / OFF),
@@ -31,7 +31,6 @@ import {
   cycleDisclaimerKey,
 } from "@/lib/cycle/dto";
 import { resolveCycleVerdict } from "@/lib/cycle/verdict";
-import { persistPredictionCache } from "@/lib/cycle/prediction-cache";
 import { addDays, dayDiff } from "@/lib/cycle/day-math";
 import { BBT_WINDOW } from "@/lib/cycle/types";
 import { DEFAULT_TIMEZONE, moodDateKey } from "@/lib/mood/date-key";
@@ -157,13 +156,14 @@ export const GET = apiHandler(async (request: NextRequest) => {
     goalAllowsFertile,
   );
 
-  // Persist the cache fire-and-forget (debounced) — never block the read.
-  let generatedAt = new Date().toISOString();
-  if (prediction) {
-    const now = new Date();
-    generatedAt = now.toISOString();
-    void persistPredictionCache(user.id, prediction, now);
-  }
+  // v1.35.3 — this read persists nothing. It used to fire
+  // `void persistPredictionCache(...)`: a write on a GET, and a discarded
+  // promise on top, so a failed write left no trace at all. The forecast the
+  // response carries is computed right here from the account's rows; the
+  // `CyclePrediction` row the reminder cron scans is written by the
+  // `cycle-prediction-refresh` job, which reaches every tracking account
+  // rather than the ones that opened this page.
+  const generatedAt = new Date().toISOString();
 
   const locale = await resolveServerLocale({
     request,