diff --git a/src/db/repositories.ts b/src/db/repositories.ts
index 397c50f511..27ae191ff5 100644
--- a/src/db/repositories.ts
+++ b/src/db/repositories.ts
@@ -2627,6 +2627,41 @@ export async function findHottestReviewTargetForRepo(
   return { targetKey: row.targetKey, count: row.count };
 }
 
+/**
+ * #review-burst-blind-spot: findHottestReviewTargetForRepo (above) only counts SUCCESSFUL publish events, so a
+ * repeat-failure retry storm (every attempt SIGKILLed / zero output, never reaching a publish) is invisible to
+ * it -- the exact incident shape c7073949 (#3747) fixed. Every AI review call's `ai_usage_events` row already
+ * carries a structured `inconclusive` boolean in its metadata_json (set at src/services/ai-review.ts's `record`
+ * call site) regardless of whether the review ever published, so this is a genuine companion signal, not a
+ * guess: the hottest PR by INCONCLUSIVE review-call count in the window, across whichever repo's calls those
+ * are. Deliberately does not touch `status` (always "ok" for a completed call, inconclusive or not) so the
+ * daily neuron-budget sum (which filters status='ok') is never affected by this query.
+ */
+export async function findHottestInconclusiveReviewTargetForRepo(
+  env: Env,
+  repoFullName: string,
+  sinceIso: string,
+): Promise<{ targetKey: string; count: number } | null> {
+  const db = getDb(env.DB);
+  const pullNumberExpr = sql<string>`json_extract(${aiUsageEvents.metadataJson}, '$.pullNumber')`;
+  const [row] = await db
+    .select({ pullNumber: pullNumberExpr, count: sql<number>`count(*)` })
+    .from(aiUsageEvents)
+    .where(
+      and(
+        eq(aiUsageEvents.feature, "ai_review_pr"),
+        gte(aiUsageEvents.createdAt, sinceIso),
+        sql`json_extract(${aiUsageEvents.metadataJson}, '$.repoFullName') = ${repoFullName}`,
+        sql`json_extract(${aiUsageEvents.metadataJson}, '$.inconclusive') = 1`,
+      ),
+    )
+    .groupBy(pullNumberExpr)
+    .orderBy(desc(sql`count(*)`))
+    .limit(1);
+  if (!row || row.pullNumber === null) return null;
+  return { targetKey: `${repoFullName}#${row.pullNumber}`, count: row.count };
+}
+
 /** Moderation-rules engine (#selfhost-mod-engine): the actor's TOTAL violation count across every rule type in
  *  `eventTypes` and EVERY repo this install tracks (no targetKey/route scoping -- `audit_events` carries no
  *  repo/installation column at all, so this is inherently install-wide, mirroring the install-wide contributor
diff --git a/src/review/ops-wire.ts b/src/review/ops-wire.ts
index 2f8e35c7c6..f218fecee2 100644
--- a/src/review/ops-wire.ts
+++ b/src/review/ops-wire.ts
@@ -30,7 +30,7 @@
 // `override_audit` D1 tables (none of which exist in gittensory's migrations yet) plus a careful soak/promote
 // design. This module is READ-ONLY observability: it reports drift; it never changes what blocks a live PR.
 
-import { findHottestReviewTargetForRepo, listRepositories } from "../db/repositories";
+import { findHottestInconclusiveReviewTargetForRepo, findHottestReviewTargetForRepo, listRepositories } from "../db/repositories";
 import { isAgentConfigured } from "../settings/autonomy";
 import { resolveRepositorySettings } from "../settings/repository-settings";
 import { loadGatePrecisionReport, type GatePrecisionReport } from "../services/gate-precision";
@@ -63,14 +63,21 @@ const REVIEW_BURST_THRESHOLD = 6;
  *  to prevent from recurring. */
 const REVIEW_BURST_WINDOW_HOURS = 2;
 
-/** One repo's outcome reports + the repo it covers — the input to the pure anomaly detector. `reviewBurst` is
- *  optional so existing snapshot-fixture tests need not be touched; absent/null means "not computed", not
- *  "healthy" -- the caller (runOpsAlerts/computeOpsStats) always populates it today. */
+/** #review-burst-blind-spot: reviewBurst (above) only sees SUCCESSFUL publishes, so a repeat-failure retry
+ *  storm (every attempt inconclusive, never publishing) sails under it indefinitely -- the exact incident
+ *  c7073949 (#3747) fixed. Lower than REVIEW_BURST_THRESHOLD on purpose: a failure burst is inherently rarer
+ *  and more anomalous than a publish burst (normal iteration never produces repeated INCONCLUSIVE calls). */
+const REVIEW_FAILURE_BURST_THRESHOLD = 3;
+
+/** One repo's outcome reports + the repo it covers — the input to the pure anomaly detector. `reviewBurst` and
+ *  `reviewFailureBurst` are optional so existing snapshot-fixture tests need not be touched; absent/null means
+ *  "not computed", not "healthy" -- the caller (runOpsAlerts/computeOpsStats) always populates both today. */
 export interface RepoOutcomeSnapshot {
   repoFullName: string;
   gatePrecision: GatePrecisionReport;
   calibration: OutcomeCalibration;
   reviewBurst?: { targetKey: string; count: number } | null | undefined;
+  reviewFailureBurst?: { targetKey: string; count: number } | null | undefined;
 }
 
 /**
@@ -120,6 +127,14 @@ export function detectOutcomeAnomalies(snapshot: RepoOutcomeSnapshot): string[]
     );
   }
 
+  // REVIEW FAILURE BURST (#review-burst-blind-spot): the publish-burst check above cannot see a repeat-failure
+  // retry storm -- every attempt produced no usable output and never reached a publish. Catch that shape too.
+  if (snapshot.reviewFailureBurst && snapshot.reviewFailureBurst.count >= REVIEW_FAILURE_BURST_THRESHOLD) {
+    out.push(
+      `review failure burst: ${snapshot.reviewFailureBurst.targetKey} produced ${snapshot.reviewFailureBurst.count} inconclusive (zero-output) AI review calls in the last ${REVIEW_BURST_WINDOW_HOURS}h with no successful publish — likely a stuck-CI finalize loop or retry storm burning tokens for no result. Investigate why this PR's reviews keep failing.`,
+    );
+  }
+
   return out;
 }
 
@@ -159,12 +174,13 @@ export async function runOpsAlerts(env: Env): Promise<Record<string, string[]>>
     const repos = await opsScanRepos(env);
     for (const repoFullName of repos) {
       try {
-        const [gatePrecision, calibration, reviewBurst] = await Promise.all([
+        const [gatePrecision, calibration, reviewBurst, reviewFailureBurst] = await Promise.all([
           loadGatePrecisionReport(env, repoFullName),
           buildRepoOutcomeCalibration(env, repoFullName),
           findHottestReviewTargetForRepo(env, repoFullName, reviewBurstSinceIso),
+          findHottestInconclusiveReviewTargetForRepo(env, repoFullName, reviewBurstSinceIso),
         ]);
-        const anomalies = detectOutcomeAnomalies({ repoFullName, gatePrecision, calibration, reviewBurst });
+        const anomalies = detectOutcomeAnomalies({ repoFullName, gatePrecision, calibration, reviewBurst, reviewFailureBurst });
         if (anomalies.length === 0) continue;
         found[repoFullName] = anomalies;
         // Structured log = gittensory's notify path (no Discord/operator webhook exists) AND the Sentry path
@@ -210,10 +226,11 @@ export async function computeOpsStats(env: Env): Promise<OpsStatsPayload> {
   const reviewBurstSinceIso = new Date(Date.now() - REVIEW_BURST_WINDOW_HOURS * 60 * 60 * 1000).toISOString();
   for (const repoFullName of repos) {
     try {
-      const [gatePrecision, calibration, reviewBurst] = await Promise.all([
+      const [gatePrecision, calibration, reviewBurst, reviewFailureBurst] = await Promise.all([
         loadGatePrecisionReport(env, repoFullName),
         buildRepoOutcomeCalibration(env, repoFullName),
         findHottestReviewTargetForRepo(env, repoFullName, reviewBurstSinceIso),
+        findHottestInconclusiveReviewTargetForRepo(env, repoFullName, reviewBurstSinceIso),
       ]);
       rows.push({
         repoFullName,
@@ -228,7 +245,7 @@ export async function computeOpsStats(env: Env): Promise<OpsStatsPayload> {
           discriminates: calibration.slop.discriminates,
         },
         recommendations: calibration.recommendations,
-        anomalies: detectOutcomeAnomalies({ repoFullName, gatePrecision, calibration, reviewBurst }),
+        anomalies: detectOutcomeAnomalies({ repoFullName, gatePrecision, calibration, reviewBurst, reviewFailureBurst }),
       });
     } catch {
       /* a per-repo failure must not blank the whole feed */
diff --git a/test/unit/ops-wire.test.ts b/test/unit/ops-wire.test.ts
index 3f47b1f9eb..282574867e 100644
--- a/test/unit/ops-wire.test.ts
+++ b/test/unit/ops-wire.test.ts
@@ -1,6 +1,6 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import { createApp } from "../../src/api/routes";
-import { recordAuditEvent, recordGateBlockOutcome, upsertPullRequestFromGitHub } from "../../src/db/repositories";
+import { recordAiUsageEvent, recordAuditEvent, recordGateBlockOutcome, upsertPullRequestFromGitHub } from "../../src/db/repositories";
 import {
   computeOpsStats,
   detectOutcomeAnomalies,
@@ -139,6 +139,22 @@ describe("detectOutcomeAnomalies — over gittensory's own outcome data", () =>
     expect(detectOutcomeAnomalies(snap).some((a) => /review burst/.test(a))).toBe(false);
   });
 
+  it("flags a review FAILURE burst — repeated inconclusive AI-review calls for the same PR with no successful publish (#review-burst-blind-spot)", () => {
+    const snap: RepoOutcomeSnapshot = { ...healthySnapshot, reviewFailureBurst: { targetKey: "owner/repo#42", count: 4 } };
+    const out = detectOutcomeAnomalies(snap);
+    expect(out.some((a) => /review failure burst/.test(a) && /owner\/repo#42/.test(a) && /4 inconclusive/.test(a))).toBe(true);
+  });
+
+  it("does NOT flag a review failure burst below the threshold", () => {
+    const snap: RepoOutcomeSnapshot = { ...healthySnapshot, reviewFailureBurst: { targetKey: "owner/repo#42", count: 2 } };
+    expect(detectOutcomeAnomalies(snap).some((a) => /review failure burst/.test(a))).toBe(false);
+  });
+
+  it("does NOT flag a review failure burst when none was computed (absent/null)", () => {
+    expect(detectOutcomeAnomalies({ ...healthySnapshot, reviewFailureBurst: null }).some((a) => /review failure burst/.test(a))).toBe(false);
+    expect(detectOutcomeAnomalies(healthySnapshot).some((a) => /review failure burst/.test(a))).toBe(false); // field omitted entirely
+  });
+
   it("does NOT flag a review burst when none was computed (absent/null)", () => {
     expect(detectOutcomeAnomalies({ ...healthySnapshot, reviewBurst: null }).some((a) => /review burst/.test(a))).toBe(false);
     expect(detectOutcomeAnomalies(healthySnapshot).some((a) => /review burst/.test(a))).toBe(false); // field omitted entirely
@@ -228,6 +244,56 @@ describe("runOpsAlerts — cron path over gittensory's outcome data", () => {
     expect(row?.anomalies.some((a) => /review burst/.test(a))).toBe(true);
   });
 
+  it("detects and reports a review FAILURE burst end-to-end -- reproduces the #3747 incident shape (repeated inconclusive calls, zero publishes) (#review-burst-blind-spot)", async () => {
+    const env = createTestEnv();
+    await seedRegisteredRepo(env, "owner/repo");
+    // Mirrors the exact incident: every AI review call for this PR came back inconclusive (zero usable output),
+    // and NONE of them ever reached a successful publish -- so findHottestReviewTargetForRepo alone sees nothing.
+    for (let i = 0; i < 4; i += 1) {
+      await recordAiUsageEvent(env, {
+        feature: "ai_review_pr",
+        model: "self-host:claude-code",
+        status: "ok",
+        estimatedNeurons: 100,
+        metadata: { repoFullName: "owner/repo", pullNumber: 99, inconclusive: true },
+      });
+    }
+    const errors = vi.spyOn(console, "error").mockImplementation(() => {});
+
+    const found = await runOpsAlerts(env);
+
+    expect(found["owner/repo"]?.some((a) => /review failure burst/.test(a) && /owner\/repo#99/.test(a) && /4 inconclusive/.test(a))).toBe(true);
+    expect(found["owner/repo"]?.some((a) => /review burst:/.test(a))).toBe(false); // the publish-only signal stays silent -- proves this is a genuinely new detector, not a duplicate.
+    const stats = await computeOpsStats(env);
+    const row = stats.repos.find((r) => r.repoFullName === "owner/repo");
+    expect(row?.anomalies.some((a) => /review failure burst/.test(a))).toBe(true);
+  });
+
+  it("does NOT flag a review failure burst from a healthy mix of successful and merely-occasional inconclusive calls", async () => {
+    const env = createTestEnv();
+    await seedRegisteredRepo(env, "owner/repo");
+    await recordAiUsageEvent(env, {
+      feature: "ai_review_pr",
+      model: "self-host:claude-code",
+      status: "ok",
+      estimatedNeurons: 100,
+      metadata: { repoFullName: "owner/repo", pullNumber: 5, inconclusive: false },
+    });
+    await recordAiUsageEvent(env, {
+      feature: "ai_review_pr",
+      model: "self-host:claude-code",
+      status: "ok",
+      estimatedNeurons: 100,
+      metadata: { repoFullName: "owner/repo", pullNumber: 5, inconclusive: true },
+    });
+    const errors = vi.spyOn(console, "error").mockImplementation(() => {});
+
+    const found = await runOpsAlerts(env);
+
+    expect(found["owner/repo"]).toBeUndefined();
+    expect(errors.mock.calls.map((c) => String(c[0])).some((line) => line.includes("ops_anomaly\""))).toBe(false);
+  });
+
   it("fails safe per-repo: a load error on one repo is logged and the scan continues (ops_anomaly_repo_error)", async () => {
     const env = createTestEnv();
     await seedRegisteredRepo(env, "owner/repo");