{
  "regression_pass": true,
  "datasets": [
    {
      "file": "ps_v1.jsonl",
      "sha256": "7b74ca1bb7f93142170e554adc9c20dea93e565c993b61142eaeeafaef32bde4",
      "tp": 10,
      "fp": 0,
      "fn": 0,
      "tn": 14,
      "category_failures": []
    },
    {
      "file": "fetch_execute_v1.jsonl",
      "sha256": "8a04b43a74a03d1d58eac0065d61e5dc3f5d3512a6044083d72814098fdd4c10",
      "tp": 8,
      "fp": 0,
      "fn": 0,
      "tn": 11,
      "category_failures": []
    },
    {
      "file": "counterexamples_v2.jsonl",
      "sha256": "82fc37b54cb3d90bbe4db72ca747d91272d864c20b4435e714543fc00c59d274",
      "tp": 3,
      "fp": 0,
      "fn": 0,
      "tn": 12,
      "category_failures": []
    },
    {
      "file": "live_gap_v1.jsonl",
      "sha256": "0d18a3e31a3fcb894926894a4898a4d138ca1fe2d4a7834f9a860f969ddec5a9",
      "tp": 0,
      "fp": 0,
      "fn": 0,
      "tn": 2,
      "category_failures": []
    },
    {
      "file": "routing_boundaries_v1.jsonl",
      "sha256": "639715fb9ebb6e755fb5028adab57f14f9b04733433b98810666bb4f93a63b62",
      "tp": 10,
      "fp": 0,
      "fn": 0,
      "tn": 16,
      "category_failures": []
    }
  ],
  "method_sha256": "8eca778f68bc889e26be2fbbe5e6bc3cadad51a53c647a7f2669e8ec554e263a",
  "case_count": 86,
  "failures": [],
  "historical": [
    {
      "file": "counterexamples_v1.jsonl",
      "sha256": "1e2a473e3aa33b449b8cc7f7db0ad7a713e79fd8ac8486fa995051468e3927b7",
      "superseded_by": "counterexamples_v2.jsonl",
      "reason": "Two historical shell labels contradicted executed second-shell behavior; preserved, not treated as current ground truth."
    }
  ],
  "independent_accuracy_evidence": false,
  "release_assessed": false
}