diff --git a/src/__tests__/latest-per-type-read-guard.test.ts b/src/__tests__/latest-per-type-read-guard.test.ts
index a0e7eb796..4ebc51836 100644
--- a/src/__tests__/latest-per-type-read-guard.test.ts
+++ b/src/__tests__/latest-per-type-read-guard.test.ts
@@ -14,6 +14,12 @@
  * the one cell; `computeSummariesSlice` is its builder and is called nowhere
  * else in production code.
  *
+ * The Prisma spelling of the same question is `measurement.groupBy` by type
+ * with nothing but `_max` (or nothing at all, a presence probe) and no time
+ * window: it groups every live row the account has to answer a question one
+ * index probe per type answers. Grouped reads that count, sum or average, or
+ * that are bounded in time, ask something else and pass.
+ *
  * Comments are stripped before matching, and the planted cases prove each
  * matcher finds what it claims to.
  */
@@ -37,6 +43,36 @@ const LATEST_PER_TYPE = /DISTINCT\s+ON\s*\(\s*(?:\w+\s*\.\s*)?"?type"?\s*\)/i;
 
 const SLICE_BUILDER_CALL = /\bcomputeSummariesSlice\s*\(/;
 
+/** Grouped presence reads that ask something the per-type probe cannot. */
+const GROUPBY_ALLOWED: Record<string, string> = {
+  "lib/measurements/import-apple-health-export.ts":
+    "Which types hold a MANUAL reading: a source-filtered presence the per-type probe does not take.",
+  "lib/jobs/insight-pregenerate.ts":
+    "Which types exist at all, with no type list to drive a per-type probe; a nightly job, not a request path.",
+};
+
+/** Each `measurement.groupBy(...)` argument that asks only latest or presence. */
+function latestOnlyGroupBys(code: string): string[] {
+  const out: string[] = [];
+  for (const m of code.matchAll(/\.measurement\s*\.\s*groupBy\s*\(/g)) {
+    let i = m.index + m[0].length;
+    let depth = 1;
+    while (depth > 0 && i < code.length) {
+      if (code[i] === "(") depth += 1;
+      else if (code[i] === ")") depth -= 1;
+      i += 1;
+    }
+    const arg = code.slice(m.index + m[0].length, i - 1);
+    const byType = /by:\s*\[\s*"type"\s*\]/.test(arg);
+    const asksMore = /\b_(?:count|sum|avg|min)\b/.test(arg);
+    const windowed = /\bmeasuredAt\s*:/.test(
+      arg.replace(/_max:\s*\{[^}]*\}/, ""),
+    );
+    if (byType && !asksMore && !windowed) out.push(arg.replace(/\s+/g, " "));
+  }
+  return out;
+}
+
 function productionFiles(): string[] {
   return walkSourceFiles(SRC, { floor: 3000 }).filter(
     (file) =>
@@ -71,6 +107,49 @@ describe("latest-per-type reads", () => {
     expect(offenders).toEqual([]);
   });
 
+  it("the groupBy matcher finds a latest-only or presence-only read", () => {
+    expect(
+      latestOnlyGroupBys(
+        `prisma.measurement.groupBy({ by: ["type"], where: { userId }, _max: { measuredAt: true } })`,
+      ),
+    ).toHaveLength(1);
+    expect(
+      latestOnlyGroupBys(
+        `prisma.measurement\n .groupBy({ by: ["type"], where: { userId, deletedAt: null } })`,
+      ),
+    ).toHaveLength(1);
+    expect(
+      latestOnlyGroupBys(
+        `prisma.measurement.groupBy({ by: ["type"], where: { userId }, _count: { _all: true }, _max: { measuredAt: true } })`,
+      ),
+    ).toEqual([]);
+    expect(
+      latestOnlyGroupBys(
+        `prisma.measurement.groupBy({ by: ["type"], where: { userId, measuredAt: { gte: since } }, _max: { measuredAt: true } })`,
+      ),
+    ).toEqual([]);
+  });
+
+  it("no production read groups every row for the latest per type", () => {
+    const files = productionFiles();
+    const grouped = files.filter((file) =>
+      /\.measurement\s*\.\s*groupBy\s*\(/.test(
+        readFileSync(join(SRC, file), "utf8"),
+      ),
+    );
+    // The tree groups measurements in a dozen places; finding none means
+    // the walk read the wrong tree.
+    expect(grouped.length).toBeGreaterThanOrEqual(8);
+    const offenders = files
+      .filter((file) => !(file in GROUPBY_ALLOWED))
+      .flatMap((file) =>
+        latestOnlyGroupBys(
+          stripComments(readFileSync(join(SRC, file), "utf8")),
+        ).map((arg) => `${file}: ${arg.slice(0, 100)}`),
+      );
+    expect(offenders).toEqual([]);
+  });
+
   it("the summaries slice is built only behind its cache cell", () => {
     const callers = productionFiles().filter((file) =>
       SLICE_BUILDER_CALL.test(
diff --git a/src/lib/daily/briefing-today-read.ts b/src/lib/daily/briefing-today-read.ts
index f9d74bb20..49ace8d14 100644
--- a/src/lib/daily/briefing-today-read.ts
+++ b/src/lib/daily/briefing-today-read.ts
@@ -11,6 +11,7 @@
  */
 import { dailyBriefingSchema, type DailyBriefing } from "@/lib/ai/schema";
 import { prisma } from "@/lib/db";
+import { readLatestEver } from "@/lib/dashboard/summary-reads";
 import type { Locale } from "@/lib/i18n/config";
 import { readBriefingGeneratedAt } from "@/lib/insights/briefing-generated-at";
 import { userDayKey } from "@/lib/tz/format";
@@ -41,21 +42,17 @@ export async function readSignalLastSeen(
   userId: string,
 ): Promise<(type: string) => string | null> {
   try {
-    const rows = await prisma.measurement.groupBy({
-      by: ["type"],
-      where: {
-        userId,
-        deletedAt: null,
-        type: {
-          in: Object.values(SIGNAL_READING_TYPE) as MeasurementType[],
-        },
-      },
-      _max: { measuredAt: true },
-    });
+    // One index probe per type for its latest live reading
+    // (`readLatestEver`), rather than a grouped max over every row.
+    const rows = await readLatestEver(
+      prisma,
+      userId,
+      Object.values(SIGNAL_READING_TYPE) as MeasurementType[],
+    );
     const byType = new Map(
       rows.map((row) => [
         String(row.type),
-        row._max.measuredAt?.toISOString() ?? null,
+        new Date(row.measured_at).toISOString(),
       ]),
     );
     return (type) => byType.get(type) ?? null;
diff --git a/src/lib/mcp/__tests__/record-search.test.ts b/src/lib/mcp/__tests__/record-search.test.ts
index 4d05457da..7a7f5e9e1 100644
--- a/src/lib/mcp/__tests__/record-search.test.ts
+++ b/src/lib/mcp/__tests__/record-search.test.ts
@@ -121,6 +121,13 @@ vi.mock("@/lib/db", () => ({
     vaccinationRecord: table("vaccinationRecord"),
   },
 }));
+
+// The search presence probe is one index lookup per type in Postgres; the
+// suite states which types hold a live reading through this stub.
+vi.mock("@/lib/dashboard/summary-reads", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("@/lib/dashboard/summary-reads")>()),
+  readLatestEver: vi.fn(async () => []),
+}));
 vi.mock("@/lib/links", () => ({
   listTargets: vi.fn(
     async (
diff --git a/src/lib/mcp/__tests__/search-fetch.test.ts b/src/lib/mcp/__tests__/search-fetch.test.ts
index d6de27350..63f09c7c3 100644
--- a/src/lib/mcp/__tests__/search-fetch.test.ts
+++ b/src/lib/mcp/__tests__/search-fetch.test.ts
@@ -42,6 +42,13 @@ vi.mock("@/lib/db", () => ({
   },
 }));
 
+// The search presence probe is one index lookup per type in Postgres; the
+// suite states which types hold a live reading through this stub.
+vi.mock("@/lib/dashboard/summary-reads", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("@/lib/dashboard/summary-reads")>()),
+  readLatestEver: vi.fn(async () => []),
+}));
+
 import { MCP_TOOLS } from "../tools";
 import { buildCoachDataInventory } from "@/lib/ai/coach/tools/inventory";
 import { executeCoachTool } from "@/lib/ai/coach/tools/executor";
diff --git a/src/lib/mcp/__tests__/tools.test.ts b/src/lib/mcp/__tests__/tools.test.ts
index f316ccfaa..f736e0d92 100644
--- a/src/lib/mcp/__tests__/tools.test.ts
+++ b/src/lib/mcp/__tests__/tools.test.ts
@@ -60,6 +60,13 @@ vi.mock("@/lib/db", () => ({
     ecgRecording: { findMany: vi.fn(async () => []) },
   },
 }));
+
+// The search presence probe is one index lookup per type in Postgres; the
+// suite states which types hold a live reading through this stub.
+vi.mock("@/lib/dashboard/summary-reads", async (importOriginal) => ({
+  ...(await importOriginal<typeof import("@/lib/dashboard/summary-reads")>()),
+  readLatestEver: vi.fn(async () => []),
+}));
 // v1.24 — the operational reads delegate to existing server-authoritative
 // engines; stub them so the registry-wide loops never reach a real engine.
 vi.mock("@/lib/medications/scheduling/next-due", () => ({
@@ -157,6 +164,7 @@ import { loadIntradayPulse } from "@/lib/analytics/intraday-pulse-io";
 import { listTargetsBySource } from "@/lib/links";
 import { decryptFromBytes } from "@/lib/ai/coach/bytes-codec";
 import type { McpAuthContext } from "../auth";
+import { readLatestEver } from "@/lib/dashboard/summary-reads";
 
 const CTX: McpAuthContext = {
   userId: "user-1",
@@ -1229,8 +1237,8 @@ describe("v1.25 clinical signals on the MCP surface", () => {
   });
 
   it("search surfaces a present clinical signal as metric:<KEY>", async () => {
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([
-      { type: "GRIP_STRENGTH" },
+    vi.mocked(readLatestEver).mockResolvedValue([
+      { type: "GRIP_STRENGTH", value: 1, measured_at: new Date() },
     ] as never);
     const result = (await tool("search").run(CTX, { query: "grip" })) as {
       results: Array<{ id: string; title: string; url: string }>;
@@ -1242,7 +1250,7 @@ describe("v1.25 clinical signals on the MCP surface", () => {
   });
 
   it("search omits a clinical signal with no recorded data", async () => {
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([] as never);
+    vi.mocked(readLatestEver).mockResolvedValue([] as never);
     const result = (await tool("search").run(CTX, { query: "grip" })) as {
       results: Array<{ id: string }>;
     };
@@ -1250,7 +1258,7 @@ describe("v1.25 clinical signals on the MCP surface", () => {
   });
 
   it("fetch hydrates a clinical signal via the baseline read (not the Coach path)", async () => {
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([] as never);
+    vi.mocked(readLatestEver).mockResolvedValue([] as never);
     vi.mocked(getMetricBaseline).mockResolvedValue({
       present: true,
       metric: "Grip strength",
@@ -1292,8 +1300,8 @@ describe("metric-status-only discovery on search/fetch (v1.30 coverage review G5
   });
 
   it("search surfaces a present metric-status-only id (undiscoverable before this wave)", async () => {
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([
-      { type: "WRIST_TEMPERATURE" },
+    vi.mocked(readLatestEver).mockResolvedValue([
+      { type: "WRIST_TEMPERATURE", value: 1, measured_at: new Date() },
     ] as never);
     const result = (await tool("search").run(CTX, { query: "wrist" })) as {
       results: Array<{ id: string; title: string }>;
@@ -1303,7 +1311,7 @@ describe("metric-status-only discovery on search/fetch (v1.30 coverage review G5
   });
 
   it("fetch hydrates a metric-status-only id via the baseline read", async () => {
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([] as never);
+    vi.mocked(readLatestEver).mockResolvedValue([] as never);
     vi.mocked(getMetricBaseline).mockResolvedValue({
       present: true,
       metric: "Wrist skin temperature",
@@ -1387,7 +1395,7 @@ describe("nutrients on search / fetch (v1.30 coverage review G1)", () => {
     } as never);
     vi.mocked(prisma.medication.findMany).mockResolvedValue([] as never);
     vi.mocked(prisma.labResult.findMany).mockResolvedValue([] as never);
-    vi.mocked(prisma.measurement.groupBy).mockResolvedValue([] as never);
+    vi.mocked(readLatestEver).mockResolvedValue([] as never);
   });
 
   it("search resolves 'water' to nutrient:water when the user has logged it", async () => {
diff --git a/src/lib/mcp/tools.ts b/src/lib/mcp/tools.ts
index d5b046a11..ee6e575f0 100644
--- a/src/lib/mcp/tools.ts
+++ b/src/lib/mcp/tools.ts
@@ -20,6 +20,7 @@
 import { z } from "zod/v4";
 
 import { prisma } from "@/lib/db";
+import { readLatestEver } from "@/lib/dashboard/summary-reads";
 import { resolveMarkerForRead } from "@/lib/labs/biomarker-resolve";
 import { annotate } from "@/lib/logging/context";
 import { getDayArgsSchema } from "@/lib/ai/coach/tools/definitions";
@@ -584,14 +585,13 @@ function searchAndFetchTools(): McpToolDefinition[] {
         // v1.41.2 — live rows only: a deleted reading is not a signal the
         // record holds, and the filter lets the probe use the live-rows
         // partial index instead of walking the tombstones.
-        const discoverablePresent = await prisma.measurement.groupBy({
-          by: ["type"],
-          where: {
-            userId: ctx.userId,
-            type: { in: discoverableSignals.map((s) => s.measurementType) },
-            deletedAt: null,
-          },
-        });
+        // A presence probe is one index lookup per type (`readLatestEver`
+        // drops a type with no live reading), not a grouping of every row.
+        const discoverablePresent = await readLatestEver(
+          prisma,
+          ctx.userId,
+          discoverableSignals.map((s) => s.measurementType),
+        );
         const presentTypes = new Set(discoverablePresent.map((r) => r.type));
         for (const sig of discoverableSignals) {
           if (!presentTypes.has(sig.measurementType)) continue;