fix: stop the gate reporting a leak when no benchmark exists

docs/reports/2026-09-27-c0-corpus-v2.md

@@ -1,7 +1,7 @@
 ---
-version: "0.1.0b"
+version: "0.1.1b"
 created_at: "2026-09-27T12:00:00+07:00,Claude,working-tree"
-last_update: "2026-09-27T12:00:00+07:00,Claude"
+last_update: "2026-09-27T13:00:00+07:00,Claude"
 status: "candidate"
 superseded_by: null
 attributes:
@@ -164,16 +164,25 @@ replayable fixture.
 4. Review the diff of `expected/*.json`. This diff is the behaviour change
    being approved.
 
-## Observation, not changed here
+## Observation, since fixed
 
-A gate verdict for a batch with no worker receipt includes the security reason
-"retrieval benchmark reported a cross-tenant leak". That reason is misleading:
-there was no benchmark at all, and the retrieval dimension already reports
-"retrieval benchmark is missing". The golden transcript records the current
-text. Correcting it is a behaviour change for a later PR.
+A gate verdict for a batch with no worker receipt used to include the security
+reason "retrieval benchmark reported a cross-tenant leak". That was misleading,
+because no benchmark existed at all. The reason text is now split in two:
+
+- no benchmark: "cross-tenant isolation is unproven: the retrieval benchmark is
+  missing."
+- a benchmark that counted leaks: "retrieval benchmark reported N
+  cross-tenant leak(s)."
+
+Both still make the security dimension a critical `FAIL`, so the gate remains
+fail-closed. The only change to the golden corpus is this reason text in
+C0.4-GENESISRAG17-RECEIPTS, which appears in the gate response and again in the
+Stage 17 evidence that carries the verdict.
 
 ## Change log
 
 | Version | Date | Status | Summary | Commit Hash | Agent |
 |---|---|---|---|---|---|
 | 0.1.0b | 2026-09-27 | candidate | C0.4 golden corpus registry v2: replayable fixtures, golden transcripts, runner, CI gate, Tier-4 readback split out | working-tree | Claude |
+| 0.1.1b | 2026-09-27 | candidate | The gate's security reason no longer reports a leak when the benchmark is missing; one case re-baselined | working-tree | Claude |

packages/gks-core/src/pipeline.mjs

@@ -651,7 +651,11 @@ export function evaluatePipelineQuality(decision, receipt, { graphReceipt = null
 
   const securityReasons = [];
   if (!decision.scope?.tenantId || !decision.scope?.portfolioId || !decision.scope?.businessId) securityReasons.push("decision scope is incomplete.");
-  if (normalizedReceipt?.benchmark?.crossTenantLeaks !== 0) securityReasons.push("retrieval benchmark reported a cross-tenant leak.");
+  // Fail closed either way, but say which: no benchmark proves the absence of a
+  // leak, which is not the same finding as a benchmark that counted one.
+  const leaks = normalizedReceipt?.benchmark?.crossTenantLeaks;
+  if (leaks === undefined) securityReasons.push("cross-tenant isolation is unproven: the retrieval benchmark is missing.");
+  else if (leaks !== 0) securityReasons.push(`retrieval benchmark reported ${leaks} cross-tenant leak(s).`);
   const security = dimension(securityReasons.length ? "FAIL" : "PASS", securityReasons.length > 0, securityReasons);
 
   const retrievalReasons = [];

tests/contract/pipeline-genesisrag17.test.mjs

@@ -367,12 +367,30 @@ describe("GenesisRAG17 pipeline contract", () => {
     const gate = await service.pipelineGate({ schemaVersion: PIPELINE_SCHEMA_VERSION, scope: batch.scope, decisionId: decision.decisionId, decisionHash: decision.decisionHash, ...worker });
     expect(gate).toMatchObject({ verdict: { verdict: "FAIL", allowPublication: false, receiptHash: null } });
     expect(gate.verdict.dimensions.graph.reasons).toContain("actual Tier4 graph receipt is missing.");
+    // No benchmark means isolation is unproven -- still a critical FAIL, but
+    // never reported as a leak that nothing measured.
+    expect(gate.verdict.dimensions.security).toEqual({ result: "FAIL", critical: true, reasons: ["cross-tenant isolation is unproven: the retrieval benchmark is missing."] });
     const evidence = (await service.pipelineEvidence({ schemaVersion: PIPELINE_SCHEMA_VERSION, scope: batch.scope, runId: batch.runId, ...auth(batch.scope) })).rows;
     expect(evidence.map((row) => row.stageNumber)).toEqual([9, 10, 11, 12, 17]);
     expect(evidence.at(-1)).toMatchObject({ stageNumber: 17, outcome: "FAILED", details: { verdict: expect.objectContaining({ verdict: "FAIL" }), publicationReceipt: null } });
     await expect(service.pipelinePublicationReceipt({ receipt: { schemaVersion: PIPELINE_SCHEMA_VERSION, scope: batch.scope, runId: batch.runId, decisionId: decision.decisionId, decisionHash: decision.decisionHash, snapshotId: "s", generation: "g", receiptHash: "a".repeat(64), publishedAt: "2026-09-07T15:00:03.000Z", pointerHash: "b".repeat(64), modelRevision: PIPELINE_MODEL.revision, transactionFrontier: "f", readback: { ok: true } }, ...worker })).rejects.toMatchObject({ code: "gks_conflict" });
   });
 
+  it("reports the measured count when the retrieval benchmark finds cross-tenant leaks", async () => {
+    const { service } = harness();
+    const batch = makeBatch({ id: "batch-leaking-benchmark" });
+    const { decision } = await submitAndClaim(service, batch);
+    const worker = auth(batch.scope, "worker");
+    const graphReceipt = graphReceiptFor(decision);
+    const graphResult = await service.pipelineGraphReceipt({ receipt: graphReceipt, ...worker });
+    const receipt = receiptFor(decision, graphResult, graphReceipt);
+    await service.pipelineWriteReceipt({ receipt: { ...receipt, benchmark: { ...receipt.benchmark, crossTenantLeaks: 2 } }, ...worker });
+    const gate = await service.pipelineGate({ schemaVersion: PIPELINE_SCHEMA_VERSION, scope: batch.scope, decisionId: decision.decisionId, decisionHash: decision.decisionHash, ...worker });
+    expect(gate.verdict).toMatchObject({ verdict: "FAIL", allowPublication: false });
+    expect(gate.verdict.dimensions.security).toEqual({ result: "FAIL", critical: true, reasons: ["retrieval benchmark reported 2 cross-tenant leak(s)."] });
+    expect(gate.verdict.dimensions.retrieval.reasons).toContain("cross-tenant leak count is non-zero.");
+  });
+
   it("treats WARN as a terminal failed gate and never publishes it", async () => {
     const { service } = harness();
     const batch = makeBatch({ id: "batch-warn-gate", scope: scope({ agentId: "agent-warn" }), entries: [{ text: "Alice and Atlas", mentions: [["Alice", "warn-alice", "Person"], ["Atlas", "warn-atlas", "Product"]] }] });

tests/fixtures/c0-qualification/expected/C0.4-GENESISRAG17-RECEIPTS.json

@@ -2971,7 +2971,7 @@
                 "result": "FAIL",
                 "critical": true,
                 "reasons": [
-                  "retrieval benchmark reported a cross-tenant leak."
+                  "cross-tenant isolation is unproven: the retrieval benchmark is missing."
                 ]
               },
               "retrieval": {
@@ -3331,7 +3331,7 @@
                       "result": "FAIL",
                       "critical": true,
                       "reasons": [
-                        "retrieval benchmark reported a cross-tenant leak."
+                        "cross-tenant isolation is unproven: the retrieval benchmark is missing."
                       ]
                     },
                     "retrieval": {

tests/fixtures/c0-qualification/registry.json

@@ -850,7 +850,7 @@
       "fixture": "cases/C0.4-GENESISRAG17-RECEIPTS.json",
       "expectedResult": "expected/C0.4-GENESISRAG17-RECEIPTS.json",
       "requestSha256": "dd696274c80ab81f0a043e99e6b3351a2f0a0fced6235a223d2862421ffb4c1f",
-      "expectedResultSha256": "ad1f8586071c87fb06b1dde3748bb84b8583d55b4863d2eaf4dce8815a363e25",
+      "expectedResultSha256": "18ebf801bbb738519e1c43142f254532fc9009f55d9fa2a8e6b1992156239191",
       "normalizedScope": {
         "portfolioId": "c0-portfolio",
         "tenantId": "c0-tenant",

tests/fixtures/c0-qualification/result-manifest.json

@@ -3,7 +3,7 @@
   "registryVersion": "c0-qualification/v2",
   "fixtureId": "gks-c0.4-golden",
   "recordedAt": "2026-09-27",
-  "recordedFrom": "f50116abdc3fe606851df4d21697027e252341c7+working-tree",
+  "recordedFrom": "29d61802d80292b342a62de5a7fa13a2ce3da378+working-tree",
   "runtime": {
     "node": "24.19.0",
     "platform": "win32",