fix: protect running validation tasks from queue cleanup

tests/test_candidate_preflight.py

@@ -391,6 +391,78 @@ class CandidatePreflightTests(unittest.TestCase):
         self.assertEqual(1, report["profileCombinationStats"][key]["failureCount"])
         self.assertEqual(1, report["recentProfileCombinationStats"][key]["consecutiveFailures"])
 
+    def test_policy_cancellation_is_excluded_from_failure_feedback(self) -> None:
+        with tempfile.TemporaryDirectory() as temporary_dir:
+            path = Path(temporary_dir) / "outcomes.jsonl"
+            tracker = OutcomeTracker(path)
+            tracker.record_submission(
+                "owner/model",
+                "gpu",
+                "vllm",
+                "text-generation",
+                "task-policy",
+                datetime.now(timezone.utc).isoformat(),
+            )
+            marked = tracker.mark_policy_cancellations(
+                [
+                    {
+                        "taskId": "task-policy",
+                        "modelId": "owner/model",
+                        "gpuType": "gpu",
+                        "cleanupReasons": ["old_model_beyond_account_queue_threshold"],
+                    }
+                ]
+            )
+            tracker.save()
+            tracker.sync_from_api(
+                TaskClient(
+                    [
+                        {
+                            "taskId": "task-policy",
+                            "status": "cancelled",
+                            "verifyResult": None,
+                        }
+                    ]
+                )  # type: ignore[arg-type]
+            )
+            report = tracker.get_stats_report()
+
+        self.assertEqual(1, marked)
+        self.assertEqual(1, report["policyCancelledRecords"])
+        self.assertEqual(0, report["terminalRecords"])
+        self.assertEqual({}, report["combinationStats"])
+        self.assertFalse(tracker.is_model_gpu_failed("owner/model", "gpu"))
+
+    def test_policy_cancellation_does_not_hide_a_racing_success(self) -> None:
+        with tempfile.TemporaryDirectory() as temporary_dir:
+            tracker = OutcomeTracker(Path(temporary_dir) / "outcomes.jsonl")
+            tracker.record_submission(
+                "owner/model",
+                "gpu",
+                "vllm",
+                "text-generation",
+                "task-success",
+                datetime.now(timezone.utc).isoformat(),
+            )
+            tracker.mark_policy_cancellations(
+                [{"taskId": "task-success", "modelId": "owner/model", "gpuType": "gpu"}]
+            )
+            tracker.sync_from_api(
+                TaskClient(
+                    [
+                        {
+                            "taskId": "task-success",
+                            "status": "success",
+                            "verifyResult": 1,
+                        }
+                    ]
+                )  # type: ignore[arg-type]
+            )
+            report = tracker.get_stats_report()
+
+        self.assertEqual(0, report["policyCancelledRecords"])
+        self.assertEqual(1, report["totals"]["successCount"])
+
     def test_failure_taxonomy_separates_platform_faults_from_model_faults(self) -> None:
         platform = classify_failure_report(
             "EXECUTE_EMPTY_RESULT",