diff --git a/src/app/api/insights/generate/__tests__/route.test.ts b/src/app/api/insights/generate/__tests__/route.test.ts
index 5ebae2f21..439d3630d 100644
--- a/src/app/api/insights/generate/__tests__/route.test.ts
+++ b/src/app/api/insights/generate/__tests__/route.test.ts
@@ -80,9 +80,16 @@ vi.mock("@/lib/logging/context", () => ({
   annotate: vi.fn(),
 }));
 
-vi.mock("@/lib/insights/features", () => ({
-  extractFeatures: vi.fn(async () => ({ stub: true })),
-}));
+vi.mock("@/lib/insights/features", async () => {
+  const actual =
+    await vi.importActual<typeof import("@/lib/insights/features")>(
+      "@/lib/insights/features",
+    );
+  return {
+    ...actual,
+    extractFeatures: vi.fn(async () => ({ stub: true })),
+  };
+});
 
 vi.mock("@/lib/insights/prompt", () => ({
   buildUserPrompt: vi.fn(() => "user"),
@@ -97,6 +104,11 @@ import { resolveProvider } from "@/lib/ai/provider";
 import { checkRateLimit } from "@/lib/rate-limit";
 import { prisma } from "@/lib/db";
 import { clearLastWorkingProviderCache } from "@/lib/ai/provider-runner";
+import {
+  extractFeatures,
+  FeaturesPayloadTooLargeError,
+} from "@/lib/insights/features";
+import { annotate } from "@/lib/logging/context";
 
 beforeEach(() => {
   vi.clearAllMocks();
@@ -344,3 +356,51 @@ describe("POST /api/insights/generate — per-status cache eviction (A7.2)", ()
     expect(prisma.auditLog.deleteMany).not.toHaveBeenCalled();
   });
 });
+
+// v1.4.36 QA H1 — when both the raw-mode extraction AND the aggregated
+// retry cross the 5 MB ceiling, the route used to throw uncaught → 500.
+// It now wraps the retry, downgrades further via the exclude filter,
+// and on a third failure returns 422 with an annotate event for ops.
+describe("POST /api/insights/generate — payload-size hard downgrade (H1)", () => {
+  it("returns 422 with an annotate event when every fallback shape is oversized", async () => {
+    makeWorkingProvider();
+    vi.mocked(prisma.user.findUnique).mockResolvedValueOnce({
+      // Raw mode so the route's first extractFeatures call is the
+      // wide shape; the retry asks for the aggregated shape.
+      insightsPrivacyMode: "raw",
+      insightsCachedAt: null,
+      insightsCachedText: null,
+      insightsExcludeMetrics: [],
+      locale: "en",
+    } as never);
+
+    // Each call mimics features.ts blowing past the ceiling. Three
+    // shots: raw, aggregated retry, hard-downgrade aggregated.
+    vi.mocked(extractFeatures).mockReset();
+    vi.mocked(extractFeatures)
+      .mockRejectedValueOnce(
+        new FeaturesPayloadTooLargeError(6_000_000, 5_242_880),
+      )
+      .mockRejectedValueOnce(
+        new FeaturesPayloadTooLargeError(5_500_000, 5_242_880),
+      )
+      .mockRejectedValueOnce(
+        new FeaturesPayloadTooLargeError(5_300_000, 5_242_880),
+      );
+
+    const res = await POST(jsonRequest({ force: true }) as never);
+    expect(res.status).toBe(422);
+    const body = (await res.json()) as ApiErrorEnvelope;
+    expect(body.error).toMatch(/payload size/i);
+
+    // The annotate event fires with insights_payload_too_large so ops
+    // can spot the regression in the wide-event pipeline.
+    const calls = vi.mocked(annotate).mock.calls;
+    const annotated = calls.find(
+      (call) =>
+        (call[0] as { meta?: Record<string, unknown> })?.meta
+          ?.insights_payload_too_large === true,
+    );
+    expect(annotated, "annotate event with insights_payload_too_large").toBeTruthy();
+  });
+});
diff --git a/src/app/api/insights/generate/route.ts b/src/app/api/insights/generate/route.ts
index 61dcc1e52..96c7e3413 100644
--- a/src/app/api/insights/generate/route.ts
+++ b/src/app/api/insights/generate/route.ts
@@ -286,13 +286,65 @@ export const POST = apiHandler(async (request: NextRequest) => {
   let features: Awaited<ReturnType<typeof extractFeatures>>;
   let payloadDowngraded = false;
   let payloadOversizeBytes: number | null = null;
+  // v1.4.36 QA H1 — extra fallback when even the aggregated shape
+  // crosses the 5 MB ceiling (very large medication history + multi-
+  // year context). Strips anthropometrics + medications + every other
+  // exclude-token-mapped block and tries one more time. On a third
+  // failure we return 422 with an annotate event for ops visibility
+  // rather than 500-ing out of an uncaught throw.
+  const MAX_DOWNGRADE_TOKENS: ReadonlyArray<string> = [
+    "anthropometrics",
+    "medications",
+    "compliance",
+    "sleep",
+    "steps",
+    "hrv",
+    "resting_hr",
+  ];
+  let payloadHardDowngraded = false;
   try {
     features = await extractFeatures(userId, includeRaw);
   } catch (err) {
     if (err instanceof FeaturesPayloadTooLargeError) {
       payloadDowngraded = true;
       payloadOversizeBytes = err.sizeBytes;
-      features = await extractFeatures(userId, false);
+      try {
+        features = await extractFeatures(userId, false);
+      } catch (retryErr) {
+        if (retryErr instanceof FeaturesPayloadTooLargeError) {
+          // Aggregated shape ALSO crossed the ceiling. Drop optional
+          // context blocks (anthropometrics, medications, sleep,
+          // steps, hrv, resting_hr) via the existing exclude filter
+          // and try once more.
+          payloadHardDowngraded = true;
+          payloadOversizeBytes = retryErr.sizeBytes;
+          try {
+            const aggregated = await extractFeatures(userId, false);
+            features = applyInsightsExcludeFilter(
+              aggregated,
+              MAX_DOWNGRADE_TOKENS,
+            );
+          } catch (finalErr) {
+            // Even the aggregated read itself blew up. Annotate and
+            // return 422 rather than let the uncaught throw 500 out.
+            annotate({
+              meta: {
+                insights_payload_too_large: true,
+                insights_payload_oversize_bytes:
+                  finalErr instanceof FeaturesPayloadTooLargeError
+                    ? finalErr.sizeBytes
+                    : payloadOversizeBytes,
+              },
+            });
+            return apiError(
+              "Your data exceeds the AI payload size limit. Reduce the data window or exclude optional blocks in Settings > AI.",
+              422,
+            );
+          }
+        } else {
+          throw retryErr;
+        }
+      }
     } else {
       throw err;
     }
@@ -315,6 +367,7 @@ export const POST = apiHandler(async (request: NextRequest) => {
       meta: {
         insights_features_downgraded: true,
         insights_features_oversize_bytes: payloadOversizeBytes,
+        insights_features_hard_downgraded: payloadHardDowngraded,
       },
     });
   }