diff --git a/migrations/0134_review_targets_cadence_idx.sql b/migrations/0134_review_targets_cadence_idx.sql
new file mode 100644
index 0000000000..4b3b0941c2
--- /dev/null
+++ b/migrations/0134_review_targets_cadence_idx.sql
@@ -0,0 +1,5 @@
+-- Supports the submission-cadence signal (#4514): getSubmitterCadence queries ALL review_targets rows (not
+-- just terminal ones -- a fresh burst of still-open submissions is exactly what this needs to catch) filtered
+-- by (project, submitter, created_at). This shape isn't covered by the existing review_targets indexes
+-- (migrations/0050), which are keyed on status/verdict/terminal_at, not created_at alongside submitter.
+CREATE INDEX IF NOT EXISTS idx_review_targets_project_submitter_created ON review_targets (project, submitter, created_at);
diff --git a/src/review/reputation-wire.ts b/src/review/reputation-wire.ts
index bb8ec25a07..04461c1287 100644
--- a/src/review/reputation-wire.ts
+++ b/src/review/reputation-wire.ts
@@ -14,7 +14,9 @@
 // the ported module degrades to "neutral" / no-op on any DB error, so this never throws into the gate.
 
 import {
+  getSubmitterCadence,
   getSubmitterReputation,
+  isMachinePacedCadence,
   recordSubmissionOutcome,
   type SubmissionOutcome,
   type SubmitterStats,
@@ -58,6 +60,12 @@ export function shouldDowngradeToDeterministic(stats: SubmitterStats): boolean {
  * downgrade to a deterministic-only review. When the flag is OFF this returns false IMMEDIATELY — no DB read —
  * so the AI-spend gate is byte-identical to today. `project` namespaces the per-(project, submitter) rows
  * (gittensory uses the repo full name). NEVER throws: the ported module already degrades to neutral on error.
+ *
+ * Also checks submission CADENCE (#4514): every quality-based signal above only tells you whether a
+ * submitter's outcomes were good or bad, never how FAST they arrived -- a fast, well-formed, strategically
+ * low-value submitter clears every quality bar while still being invisible to those signals. A cadence read
+ * this tight, sustained across this many consecutive submissions, is not a pattern any human contributor
+ * plausibly sustains, independent of whether the submissions themselves look fine.
  */
 export async function shouldSkipAiForReputation(
   env: Env,
@@ -65,7 +73,9 @@ export async function shouldSkipAiForReputation(
 ): Promise<boolean> {
   if (!isReputationEnabled(env)) return false;
   const stats = await getSubmitterReputation(env, args.project, args.submitter ?? undefined);
-  return shouldDowngradeToDeterministic(stats);
+  if (shouldDowngradeToDeterministic(stats)) return true;
+  const cadence = await getSubmitterCadence(env, args.project, args.submitter ?? undefined);
+  return isMachinePacedCadence(cadence);
 }
 
 /**
diff --git a/src/review/submitter-reputation.ts b/src/review/submitter-reputation.ts
index 72558c8615..04b9c5f19e 100644
--- a/src/review/submitter-reputation.ts
+++ b/src/review/submitter-reputation.ts
@@ -70,6 +70,58 @@ export const REPUTATION_WINDOW_DAYS = 90;
 // Hard ceiling on rows pulled for one submitter's window so a pathological history can't blow the query up.
 const REPUTATION_WINDOW_ROW_CAP = 500;
 
+// ── Submission-cadence signal (#4514). Every signal above is QUALITY-based (was the outcome good or bad) --
+// none of them have a TIMING dimension, so a fast, well-formed, strategically-low-value submitter is
+// invisible to the one dimension (superhuman pace) that would otherwise be a strong tell. This is queried
+// from ALL review_targets rows (not just terminal ones, unlike the quality signal above) -- a fresh burst of
+// still-open submissions is exactly the case this needs to catch, and by the time they become terminal the
+// (paid) AI review has already run on each one. ──
+const CADENCE_WINDOW_HOURS = 24;
+// Need at least this many recent submissions before judging pace at all -- a lone fast submission (a real
+// contributor who happened to open two PRs close together) is not a pattern.
+const CADENCE_MIN_SAMPLE = 5;
+// A human contributor, even a fast one, does not sustain a sub-10-minute median gap between distinct PR
+// submissions across many consecutive attempts -- reading, writing, and testing each change takes real time.
+const CADENCE_MAX_MEDIAN_GAP_MS = 10 * 60 * 1000;
+
+export type SubmissionCadence = { count: number; medianGapMs: number | null };
+
+/** Pure: the median gap (ms) between consecutive submissions, given their created_at timestamps in any order.
+ *  `medianGapMs` is `null` when there are fewer than 2 samples (no gap to measure). */
+export function computeSubmissionCadence(createdAtIsoTimestamps: readonly string[]): SubmissionCadence {
+  const sorted = [...createdAtIsoTimestamps].map((t) => new Date(t).getTime()).sort((a, b) => a - b);
+  if (sorted.length < 2) return { count: sorted.length, medianGapMs: null };
+  const gaps: number[] = [];
+  for (let i = 1; i < sorted.length; i++) gaps.push(sorted[i]! - sorted[i - 1]!);
+  gaps.sort((a, b) => a - b);
+  const mid = Math.floor(gaps.length / 2);
+  const medianGapMs = gaps.length % 2 === 0 ? (gaps[mid - 1]! + gaps[mid]!) / 2 : gaps[mid]!;
+  return { count: sorted.length, medianGapMs };
+}
+
+/** Pure: does this cadence read as machine-paced? Needs both a real sample size AND a gap tighter than any
+ *  human contributor plausibly sustains across that many consecutive attempts. */
+export function isMachinePacedCadence(cadence: SubmissionCadence): boolean {
+  return cadence.count >= CADENCE_MIN_SAMPLE && cadence.medianGapMs !== null && cadence.medianGapMs < CADENCE_MAX_MEDIAN_GAP_MS;
+}
+
+/** Per-repo submission cadence for one submitter over the last {@link CADENCE_WINDOW_HOURS}. Fail-safe:
+ *  any read error degrades to `{ count: 0, medianGapMs: null }` (never machine-paced), identical in spirit to
+ *  {@link getSubmitterReputation}'s fail-safe-to-neutral. */
+export async function getSubmitterCadence(env: Env, project: string, submitter: string | undefined): Promise<SubmissionCadence> {
+  if (!submitter) return { count: 0, medianGapMs: null };
+  try {
+    const result = await storage(env)
+      .prepare(`SELECT created_at AS createdAt FROM review_targets WHERE project = ? AND submitter = ? AND created_at >= datetime('now', ?) ORDER BY created_at DESC LIMIT ?`)
+      .bind(project, submitter, `-${CADENCE_WINDOW_HOURS} hours`, REPUTATION_WINDOW_ROW_CAP)
+      .all<{ createdAt: string }>();
+    const createdAts = (result?.results ?? []).map((r) => r.createdAt);
+    return computeSubmissionCadence(createdAts);
+  } catch {
+    return { count: 0, medianGapMs: null };
+  }
+}
+
 // ── reasonCode → quality bucket (#reputation-redesign). Buckets reflect the LIVE D1 reasonCode taxonomy. ──
 //   SUCCESS: a genuine reviewer/merge approval.
 //   QUALITY_FAIL: a genuine RECENT reviewer reject (real quality signal).
diff --git a/test/unit/reputation-wiring.test.ts b/test/unit/reputation-wiring.test.ts
index 9120da5be0..e41b76e754 100644
--- a/test/unit/reputation-wiring.test.ts
+++ b/test/unit/reputation-wiring.test.ts
@@ -149,6 +149,56 @@ describe("shouldSkipAiForReputation (helper)", () => {
     expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: "burster" })).toBe(true);
     expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: "newcomer" })).toBe(false);
   });
+
+  it("FLAG-ON: false for a null submitter (the ?? undefined coalesce on both the quality and cadence reads)", async () => {
+    const env = createTestEnv({ GITTENSORY_REVIEW_REPUTATION: "true" });
+    expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: null })).toBe(false);
+  });
+
+  describe("submission-cadence signal (#4514)", () => {
+    async function seedReviewTarget(env: Env, args: { number: number; submitter: string; createdAt: string }) {
+      await env.DB.prepare(
+        `INSERT INTO review_targets (id, project, kind, repo, number, submitter, status, decision_json, terminal_at, created_at)
+         VALUES (?, 'acme/widgets', 'pull_request', 'acme/widgets', ?, ?, 'merged', ?, ?, ?)`,
+      )
+        .bind(`acme/widgets:pull_request:acme/widgets#${args.number}`, args.number, args.submitter, JSON.stringify({ reasonCode: "dual_review_approved" }), args.createdAt, args.createdAt)
+        .run();
+    }
+
+    it("FLAG-ON: true for a machine-paced submitter even though every submission itself looks fine (quality-neutral)", async () => {
+      const env = createTestEnv({ GITTENSORY_REVIEW_REPUTATION: "true" });
+      // Anchored to now (minus a couple hours of headroom) -- the cadence query only looks back 24h, so a
+      // fixed past date would fall outside the window and vacuously read as "0 samples, not machine-paced".
+      const t0 = Date.now() - 2 * 60 * 60_000;
+      for (let i = 0; i < 5; i++) {
+        // All merged/approved -- the QUALITY signal alone stays neutral/trusted; only cadence should trip this.
+        await seedReviewTarget(env, { number: i, submitter: "speedster", createdAt: new Date(t0 + i * 5 * 60_000).toISOString() });
+      }
+      expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: "speedster" })).toBe(true);
+    });
+
+    it("FLAG-ON: false for the same number of submissions spread naturally over hours (comfortably human pace)", async () => {
+      const env = createTestEnv({ GITTENSORY_REVIEW_REPUTATION: "true" });
+      const t0 = Date.now() - 20 * 60 * 60_000;
+      for (let i = 0; i < 5; i++) {
+        await seedReviewTarget(env, { number: i + 100, submitter: "steady", createdAt: new Date(t0 + i * 3 * 60 * 60_000).toISOString() });
+      }
+      expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: "steady" })).toBe(false);
+    });
+
+    it("FLAG-ON: skips the (extra) cadence read once the quality/burst signal already justifies downgrading", async () => {
+      const env = createTestEnv({ GITTENSORY_REVIEW_REPUTATION: "true" });
+      await seedSubmitter(env, { project: "acme/widgets", submitter: "burster", submissions: 12, merged: 0, closed: 12, manual: 0 });
+      const spy = vi.spyOn(env.DB, "prepare");
+      const before = spy.mock.calls.length;
+      expect(await shouldSkipAiForReputation(env, { project: "acme/widgets", submitter: "burster" })).toBe(true);
+      // Exactly the calls the quality/burst read itself makes (submitter_stats + review_targets window) --
+      // no additional prepare() for a cadence query once the burst check alone already returned true.
+      const afterQualityOnlyCallCount = spy.mock.calls.length - before;
+      spy.mockRestore();
+      expect(afterQualityOnlyCallCount).toBe(2);
+    });
+  });
 });
 
 describe("processGitHubWebhook records the reputation outcome on a terminal PR (flag-ON call site)", () => {
diff --git a/test/unit/submitter-reputation.test.ts b/test/unit/submitter-reputation.test.ts
index 80b72aa8f0..6a18a6a8bc 100644
--- a/test/unit/submitter-reputation.test.ts
+++ b/test/unit/submitter-reputation.test.ts
@@ -1,9 +1,12 @@
 import { describe, expect, it } from "vitest";
 import {
   classifyOutcome,
+  computeSubmissionCadence,
   countOutcomes,
   DEFAULT_REPUTATION_CONFIG,
+  getSubmitterCadence,
   getSubmitterReputation,
+  isMachinePacedCadence,
   recordSubmissionOutcome,
   REPUTATION_WINDOW_DAYS,
   type ReputationConfig,
@@ -273,6 +276,104 @@ describe("recordSubmissionOutcome / getSubmitterReputation (D1, fail-safe)", ()
   });
 });
 
+describe("computeSubmissionCadence (pure) (#4514)", () => {
+  it("returns count and null medianGapMs for 0 or 1 samples (nothing to measure a gap between)", () => {
+    expect(computeSubmissionCadence([])).toEqual({ count: 0, medianGapMs: null });
+    expect(computeSubmissionCadence(["2026-01-01T00:00:00.000Z"])).toEqual({ count: 1, medianGapMs: null });
+  });
+
+  it("computes the median gap between consecutive submissions, order-independent", () => {
+    // Gaps: 10min, 20min, 30min -> sorted [10,20,30] -> median 20min.
+    const t0 = new Date("2026-01-01T00:00:00.000Z").getTime();
+    const timestamps = [t0, t0 + 10 * 60_000, t0 + 30 * 60_000, t0 + 60 * 60_000].map((ms) => new Date(ms).toISOString());
+    // Shuffle the input order -- the function must sort internally, not assume caller ordering.
+    const shuffled = [timestamps[2]!, timestamps[0]!, timestamps[3]!, timestamps[1]!];
+    expect(computeSubmissionCadence(shuffled)).toEqual({ count: 4, medianGapMs: 20 * 60_000 });
+  });
+
+  it("averages the two middle gaps for an even number of gaps", () => {
+    // 3 timestamps -> 2 gaps: 10min, 30min -> even count -> average = 20min.
+    const t0 = new Date("2026-01-01T00:00:00.000Z").getTime();
+    const timestamps = [t0, t0 + 10 * 60_000, t0 + 40 * 60_000].map((ms) => new Date(ms).toISOString());
+    expect(computeSubmissionCadence(timestamps)).toEqual({ count: 3, medianGapMs: 20 * 60_000 });
+  });
+});
+
+describe("isMachinePacedCadence (pure) (#4514)", () => {
+  it("requires BOTH the minimum sample size AND a sub-threshold median gap", () => {
+    // Below minSample (5) -- fast, but not enough samples to call it a pattern.
+    expect(isMachinePacedCadence({ count: 4, medianGapMs: 60_000 })).toBe(false);
+    // Enough samples, but the gap is comfortably human (well over 10min).
+    expect(isMachinePacedCadence({ count: 10, medianGapMs: 60 * 60_000 })).toBe(false);
+    // No gap at all to measure (count < 2 internally, or explicitly null).
+    expect(isMachinePacedCadence({ count: 8, medianGapMs: null })).toBe(false);
+    // Enough samples AND a tight gap -- machine-paced.
+    expect(isMachinePacedCadence({ count: 5, medianGapMs: 5 * 60_000 })).toBe(true);
+    expect(isMachinePacedCadence({ count: 20, medianGapMs: 60_000 })).toBe(true);
+  });
+
+  it("is a boundary at exactly the configured thresholds", () => {
+    // Exactly at minSample, exactly under the max gap -- still counts.
+    expect(isMachinePacedCadence({ count: 5, medianGapMs: 10 * 60_000 - 1 })).toBe(true);
+    // Exactly AT the max gap -- not strictly under, so not machine-paced.
+    expect(isMachinePacedCadence({ count: 5, medianGapMs: 10 * 60_000 })).toBe(false);
+  });
+});
+
+describe("getSubmitterCadence (D1, fail-safe) (#4514)", () => {
+  function makeCadenceEnv(createdAts: string[]): Env {
+    return {
+      DB: {
+        prepare: () => ({
+          bind: () => ({
+            all: async () => ({ results: createdAts.map((createdAt) => ({ createdAt })) }),
+          }),
+        }),
+      },
+    } as unknown as Env;
+  }
+
+  it("returns count 0 / null with no submitter (early return, no DB touch)", async () => {
+    expect(await getSubmitterCadence({} as Env, "p", undefined)).toEqual({ count: 0, medianGapMs: null });
+  });
+
+  it("derives cadence from the queried created_at timestamps", async () => {
+    const t0 = new Date("2026-01-01T00:00:00.000Z").getTime();
+    const env = makeCadenceEnv([t0, t0 + 5 * 60_000, t0 + 10 * 60_000, t0 + 15 * 60_000, t0 + 20 * 60_000].map((ms) => new Date(ms).toISOString()));
+    const cadence = await getSubmitterCadence(env, "p", "farmer99");
+    expect(cadence).toEqual({ count: 5, medianGapMs: 5 * 60_000 });
+    expect(isMachinePacedCadence(cadence)).toBe(true);
+  });
+
+  it("fail-safe: degrades to count 0 / null when the query throws, never throws into the caller", async () => {
+    const env = {
+      DB: {
+        prepare: () => ({
+          bind: () => ({
+            all: async () => {
+              throw new Error("D1 boom");
+            },
+          }),
+        }),
+      },
+    } as unknown as Env;
+    expect(await getSubmitterCadence(env, "p", "farmer99")).toEqual({ count: 0, medianGapMs: null });
+  });
+
+  it("fail-safe: degrades to count 0 / null when the query returns a malformed result (?? [] fallback)", async () => {
+    const env = {
+      DB: {
+        prepare: () => ({
+          bind: () => ({
+            all: async () => undefined,
+          }),
+        }),
+      },
+    } as unknown as Env;
+    expect(await getSubmitterCadence(env, "p", "farmer99")).toEqual({ count: 0, medianGapMs: null });
+  });
+});
+
 // A minimal D1 stub: the first query (.first) returns submitter_stats; the window query (.all) returns the
 // review_targets rows. Both come off the same prepared-statement stub (the two call sites use .first vs .all).
 function makeEnv(opts: { statRow: { submissions: number; merged: number; closed: number; manual: number } | null; windowRows: Row[] }): Env {