diff --git a/packages/loopover-engine/src/finding-severity-calibration.ts b/packages/loopover-engine/src/finding-severity-calibration.ts
index 443a4e4a55..5b292eb771 100644
--- a/packages/loopover-engine/src/finding-severity-calibration.ts
+++ b/packages/loopover-engine/src/finding-severity-calibration.ts
@@ -363,7 +363,10 @@ function normalizeCompositeWeights(weights: FindingSeverityCalibrationWeights |
     ),
   };
   const total = raw.objectiveAnchor + raw.pairwiseJudge + raw.structuredFindingSeverity;
-  if (total <= 0) return DEFAULT_COMPOSITE_WEIGHTS;
+  // Preserve explicitly-zeroed weights rather than substituting the defaults: a caller that zeroes every component
+  // must reach the objective-only fallback in the composite scorer, not silently get the default 45/35/20 blend
+  // (converges with reviewer-consensus-calibration.ts's already-correct behavior; #6170).
+  if (total <= 0) return { objectiveAnchor: 0, pairwiseJudge: 0, structuredFindingSeverity: 0 };
   return {
     objectiveAnchor: raw.objectiveAnchor / total,
     pairwiseJudge: raw.pairwiseJudge / total,
diff --git a/packages/loopover-engine/src/gate-verdict-calibration.ts b/packages/loopover-engine/src/gate-verdict-calibration.ts
index d644dc0ad2..c0569ebe68 100644
--- a/packages/loopover-engine/src/gate-verdict-calibration.ts
+++ b/packages/loopover-engine/src/gate-verdict-calibration.ts
@@ -283,7 +283,8 @@ function sanitizeGateVerdictCalibrationIngestion(
 
   for (const row of ingestion.rejected) {
     if (!isRecord(row)) continue;
-    const repoFullName = typeof row.repoFullName === "string" ? normalizeRepoFullName(row.repoFullName) : null;
+    const repoFullName =
+      typeof row.repoFullName === "string" ? (normalizeRepoFullName(row.repoFullName) ?? normalizeId(row.repoFullName)) : null;
     const replayRunId = typeof row.replayRunId === "string" ? normalizeId(row.replayRunId) : null;
     const gateRunId = typeof row.gateRunId === "string" ? normalizeId(row.gateRunId) : null;
     const reason = row.reason;
@@ -315,7 +316,10 @@ function normalizeCompositeWeights(weights: GateVerdictCalibrationWeights | unde
     ),
   };
   const total = raw.objectiveAnchor + raw.pairwiseJudge + raw.structuredGateVerdict;
-  if (total <= 0) return DEFAULT_COMPOSITE_WEIGHTS;
+  // Preserve explicitly-zeroed weights rather than substituting the defaults: a caller that zeroes every component
+  // must reach the objective-only fallback in the composite scorer, not silently get the default 45/35/20 blend
+  // (converges with reviewer-consensus-calibration.ts's already-correct behavior; #6170).
+  if (total <= 0) return { objectiveAnchor: 0, pairwiseJudge: 0, structuredGateVerdict: 0 };
   return {
     objectiveAnchor: raw.objectiveAnchor / total,
     pairwiseJudge: raw.pairwiseJudge / total,
diff --git a/packages/loopover-engine/test/finding-severity-calibration.test.ts b/packages/loopover-engine/test/finding-severity-calibration.test.ts
index 0d8e4485ac..ee5d1f07c5 100644
--- a/packages/loopover-engine/test/finding-severity-calibration.test.ts
+++ b/packages/loopover-engine/test/finding-severity-calibration.test.ts
@@ -470,3 +470,16 @@ test("renderAuditMarkdown tolerates malformed pre-ingested rejected rows after s
   assert.deepEqual(result.audit.rejected, []);
   assert.doesNotThrow(() => renderFindingSeverityCalibrationAuditMarkdown(result));
 });
+
+test("computeFindingSeverityCompositeCalibrationScore falls back to objective-only when all weights are explicitly zero (#6170)", () => {
+  const result = computeFindingSeverityCompositeCalibrationScore({
+    objectiveAnchor: 0.4,
+    pairwise: 0.4,
+    findingSeverity: [signal()],
+    weights: { objectiveAnchor: 0, pairwiseJudge: 0, structuredFindingSeverity: 0 },
+  });
+  // Explicitly zeroing every component falls back to objective-only -- NOT the default 45/35/20 blend
+  // (converges with reviewer-consensus-calibration.ts).
+  assert.deepEqual(result.weights, { objectiveAnchor: 1, pairwiseJudge: 0, structuredFindingSeverity: 0 });
+  assert.equal(result.compositeScore, 0.4);
+});
diff --git a/packages/loopover-engine/test/gate-verdict-calibration.test.ts b/packages/loopover-engine/test/gate-verdict-calibration.test.ts
index 2d18452a34..60c995e75b 100644
--- a/packages/loopover-engine/test/gate-verdict-calibration.test.ts
+++ b/packages/loopover-engine/test/gate-verdict-calibration.test.ts
@@ -580,3 +580,44 @@ test("computeGateVerdictCompositeCalibrationScore ignores malformed pre-ingested
   assert.deepEqual(result.audit.contributingRepos, []);
   assert.deepEqual(result.audit.rejected, []);
 });
+
+test("computeGateVerdictCompositeCalibrationScore falls back to objective-only when all weights are explicitly zero (#6170)", () => {
+  const result = computeGateVerdictCompositeCalibrationScore({
+    objectiveAnchor: 0.4,
+    pairwise: 0.4,
+    gateVerdicts: [
+      {
+        repoFullName: "acme/widgets",
+        replayRunId: "replay-1",
+        gateRunId: "gate-1",
+        optedIn: true,
+        dimensions: [{ dimension: "correctness", outcome: "pass" }],
+      },
+    ],
+    weights: { objectiveAnchor: 0, pairwiseJudge: 0, structuredGateVerdict: 0 },
+  });
+  // Explicitly zeroing every component falls back to objective-only -- NOT the default 45/35/20 blend
+  // (converges with reviewer-consensus-calibration.ts).
+  assert.deepEqual(result.weights, { objectiveAnchor: 1, pairwiseJudge: 0, structuredGateVerdict: 0 });
+  assert.equal(result.compositeScore, 0.4);
+});
+
+test("computeGateVerdictCompositeCalibrationScore preserves a malformed-repo (invalid_repo) rejected row instead of dropping it (#6170)", () => {
+  const result = computeGateVerdictCompositeCalibrationScore({
+    objectiveAnchor: 0.5,
+    pairwise: 0.5,
+    gateVerdicts: {
+      accepted: [],
+      rejected: [
+        { repoFullName: "acme/widgets", replayRunId: "replay-1", gateRunId: "gate-1", reason: "not_opted_in" },
+        { repoFullName: "bad", replayRunId: "replay-2", gateRunId: "gate-2", reason: "invalid_repo" },
+      ],
+    } as never,
+  });
+  // "bad" is not a valid owner/repo (normalizeRepoFullName -> null); the `?? normalizeId` fallback preserves the
+  // raw string instead of dropping the row, matching finding-severity + reviewer-consensus.
+  assert.deepEqual(result.audit.rejected, [
+    { repoFullName: "acme/widgets", replayRunId: "replay-1", gateRunId: "gate-1", reason: "not_opted_in" },
+    { repoFullName: "bad", replayRunId: "replay-2", gateRunId: "gate-2", reason: "invalid_repo" },
+  ]);
+});
diff --git a/test/unit/engine-calibration-convergence.test.ts b/test/unit/engine-calibration-convergence.test.ts
new file mode 100644
index 0000000000..685da61e3d
--- /dev/null
+++ b/test/unit/engine-calibration-convergence.test.ts
@@ -0,0 +1,64 @@
+import { describe, expect, it } from "vitest";
+import {
+  computeGateVerdictCompositeCalibrationScore,
+  computeFindingSeverityCompositeCalibrationScore,
+} from "../../packages/loopover-engine/src/index";
+
+// Converges gate-verdict + finding-severity calibration with reviewer-consensus-calibration.ts's already-correct
+// all-zero-weight + malformed-repo handling (#6170). These run under vitest (Codecov-measured) so the changed
+// engine-src branches are covered; the engine's own node:test suites carry the equivalent assertions.
+describe("gate-verdict/finding-severity calibration convergence (#6170)", () => {
+  it("gate-verdict: all-zero weights fall back to objective-only, not the default 45/35/20 blend", () => {
+    const result = computeGateVerdictCompositeCalibrationScore({
+      objectiveAnchor: 0.4,
+      pairwise: 0.4,
+      gateVerdicts: [
+        {
+          repoFullName: "acme/widgets",
+          replayRunId: "replay-1",
+          gateRunId: "gate-1",
+          optedIn: true,
+          dimensions: [{ dimension: "correctness", outcome: "pass" }],
+        },
+      ],
+      weights: { objectiveAnchor: 0, pairwiseJudge: 0, structuredGateVerdict: 0 },
+    });
+    expect(result.weights).toEqual({ objectiveAnchor: 1, pairwiseJudge: 0, structuredGateVerdict: 0 });
+    expect(result.compositeScore).toBe(0.4);
+  });
+
+  it("finding-severity: all-zero weights fall back to objective-only, not the default blend", () => {
+    const result = computeFindingSeverityCompositeCalibrationScore({
+      objectiveAnchor: 0.4,
+      pairwise: 0.4,
+      findingSeverity: [
+        { repoFullName: "acme/widgets", replayRunId: "replay-1", reviewRunId: "review-1", optedIn: true, tiers: [{ tier: "blocker", total: 2, confirmed: 2 }] },
+      ],
+      weights: { objectiveAnchor: 0, pairwiseJudge: 0, structuredFindingSeverity: 0 },
+    });
+    expect(result.weights).toEqual({ objectiveAnchor: 1, pairwiseJudge: 0, structuredFindingSeverity: 0 });
+    expect(result.compositeScore).toBe(0.4);
+  });
+
+  it("gate-verdict: preserves a malformed-repo (invalid_repo) rejected row instead of dropping it; keeps a valid repo and drops a non-string one", () => {
+    const result = computeGateVerdictCompositeCalibrationScore({
+      objectiveAnchor: 0.5,
+      pairwise: 0.5,
+      gateVerdicts: {
+        accepted: [],
+        rejected: [
+          { repoFullName: "acme/widgets", replayRunId: "replay-1", gateRunId: "gate-1", reason: "not_opted_in" },
+          { repoFullName: "bad", replayRunId: "replay-2", gateRunId: "gate-2", reason: "invalid_repo" },
+          { repoFullName: 123, replayRunId: "replay-3", gateRunId: "gate-3", reason: "invalid_repo" },
+        ],
+      } as never,
+    });
+    // "bad" is not a valid owner/repo, so normalizeRepoFullName returns null; the `?? normalizeId` fallback
+    // preserves the raw string instead of dropping the row (matching reviewer-consensus). The non-string repo
+    // (123) still drops (ternary false branch).
+    expect(result.audit.rejected).toEqual([
+      { repoFullName: "acme/widgets", replayRunId: "replay-1", gateRunId: "gate-1", reason: "not_opted_in" },
+      { repoFullName: "bad", replayRunId: "replay-2", gateRunId: "gate-2", reason: "invalid_repo" },
+    ]);
+  });
+});