{
  "scope": "LOCAL APPLICATION VERIFICATION ONLY",
  "live_api_calls": 0,
  "synthetic_cases": 27,
  "unit_and_fault_tests": 18,
  "task_runs": {
    "classification": {
      "mode": "demo",
      "evidence": "synthetic fixtures; NOT model performance",
      "task": "classification",
      "created_at": "2026-09-20T07:15:57.462993+00:00",
      "requested_model": "jev-latest",
      "returned_models": [
        "synthetic-fixture-not-jev"
      ],
      "dataset_sha256": "484631cd10159ede0af0e8fb8fd582f0459eba4019d9f019dfade7878121fa5f",
      "rubric_sha256": "e1e765c46bfb04e10f3254fb2810d9178c46689cb1c4af58de6ec744d7550258",
      "thresholds": {
        "confidence": 0.7,
        "low": 0.2,
        "high": 0.8
      },
      "input_rows": 6,
      "failed_rows": 0,
      "metrics": {
        "exact_policy_matches": 6,
        "valid_rows": 6,
        "exact_policy_match_rate": 1.0,
        "review_count": 2
      }
    },
    "context-filtering": {
      "mode": "demo",
      "evidence": "synthetic fixtures; NOT model performance",
      "task": "context-filtering",
      "created_at": "2026-09-20T07:15:57.565537+00:00",
      "requested_model": "jev-latest",
      "returned_models": [
        "synthetic-fixture-not-jev"
      ],
      "dataset_sha256": "b831423b132071bdea580054dc318345cff0e93a0cc1bcb11f10631ba21e6a1a",
      "rubric_sha256": "c5506bbeb26824c59b3f295f76471aabc499997abdb7fad3759317311664f95b",
      "thresholds": {
        "confidence": 0.7,
        "low": 0.2,
        "high": 0.8
      },
      "input_rows": 5,
      "failed_rows": 0,
      "metrics": {
        "exact_policy_matches": 5,
        "valid_rows": 5,
        "exact_policy_match_rate": 1.0,
        "review_count": 0
      }
    },
    "moderation": {
      "mode": "demo",
      "evidence": "synthetic fixtures; NOT model performance",
      "task": "moderation",
      "created_at": "2026-09-20T07:15:57.643800+00:00",
      "requested_model": "jev-latest",
      "returned_models": [
        "synthetic-fixture-not-jev"
      ],
      "dataset_sha256": "7571a6109c171d28f733f1242dca12dc386364ada4d43d13a5618251a03ed730",
      "rubric_sha256": "5ed21298861f53cb350a3b02fc4aa85d618c4c49f0496249d040768a890755c8",
      "thresholds": {
        "confidence": 0.7,
        "low": 0.2,
        "high": 0.8
      },
      "input_rows": 6,
      "failed_rows": 0,
      "metrics": {
        "exact_policy_matches": 6,
        "valid_rows": 6,
        "exact_policy_match_rate": 1.0,
        "review_count": 1
      }
    },
    "reranking": {
      "mode": "demo",
      "evidence": "synthetic fixtures; NOT model performance",
      "task": "reranking",
      "created_at": "2026-09-20T07:15:57.716596+00:00",
      "requested_model": "jev-latest",
      "returned_models": [
        "synthetic-fixture-not-jev"
      ],
      "dataset_sha256": "80fc08b6c7dc541af1db5f008f03e2c918ef48c2360f933a34e6773a460a5f70",
      "rubric_sha256": "0d6ce973359660ebf675d48aaf31a8d26d5d61b8c3496d5045827d026058ef02",
      "thresholds": {
        "confidence": 0.7,
        "low": 0.2,
        "high": 0.8
      },
      "input_rows": 4,
      "failed_rows": 0,
      "metrics": {
        "k": 2,
        "recall_at_k": 0.6666666666666666,
        "ndcg_at_k": 1.0
      }
    },
    "tool-routing": {
      "mode": "demo",
      "evidence": "synthetic fixtures; NOT model performance",
      "task": "tool-routing",
      "created_at": "2026-09-20T07:15:57.865203+00:00",
      "requested_model": "jev-latest",
      "returned_models": [
        "synthetic-fixture-not-jev"
      ],
      "dataset_sha256": "b90899808f5aea8cc5c0c17b9c764518a0d8fa883c165523f69a86eba27cd29b",
      "rubric_sha256": "6c4eb240edce9f24863f1b6e09e9c5bea8ae89e4a1ecc48f1e3dd93d9abf4594",
      "thresholds": {
        "confidence": 0.7,
        "low": 0.2,
        "high": 0.8
      },
      "input_rows": 6,
      "failed_rows": 0,
      "metrics": {
        "exact_policy_matches": 6,
        "valid_rows": 6,
        "exact_policy_match_rate": 1.0,
        "review_count": 3
      }
    }
  },
  "limitations": [
    "No TypeSafe API credential available",
    "Synthetic response fixtures do not measure model quality",
    "Network retries tested with injected HTTP exceptions, not provider outages",
    "Videos replay actual local command output with annotations; no live provider footage"
  ],
  "threshold_counterfactual": {
    "task": "tool-routing",
    "case": "overlap",
    "cutoff_0_7": "review",
    "cutoff_0_2": "shipping",
    "source": "synthetic fixture",
    "verified": true
  }
}