{
  "run_id": "decider__shell_risk_safe__20261011T210654Z",
  "backend": "decider",
  "model_version": "StrandsAgents/strands-decider-2B-hobson-v19@strands-decider-0.1.0",
  "suite_id": "shell_risk_safe",
  "suite_version": "1.0.0",
  "suite_title": "Shell auto-run safety (synthetic)",
  "primitive": "noul",
  "protocol": "1.0.0",
  "split": "test",
  "limited": true,
  "limit": 200,
  "prepare_ms": 0.0,
  "latency_p50_ms": 1982.4083944995436,
  "latency_p95_ms": 2452.5846836007076,
  "latency_mean_ms": 2023.7347900800523,
  "batch_size": 1,
  "concurrency": 1,
  "cost_usd_total": 0.0,
  "cost_per_1k_usd": 0.0,
  "hardware": {
    "platform": "Linux-6.12.94+-x86_64-with-glibc2.41",
    "python": "3.12.14",
    "machine": "x86_64",
    "processor": "x86_64",
    "cpu_count": 8
  },
  "robustness_probes": {
    "order_reverse": {
      "n": 53,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "order_shuffle": {
      "n": 53,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "paraphrase": {
      "n": 53,
      "invalid_pairs": 0,
      "flip_rate": 0.7358490566037735,
      "probability_drift": 0.21908490566037736,
      "attack_success_rate": null
    },
    "irrelevant_context": {
      "n": 53,
      "invalid_pairs": 0,
      "flip_rate": 0.03773584905660377,
      "probability_drift": 0.02269999999999999,
      "attack_success_rate": null
    },
    "adversarial": {
      "n": 53,
      "invalid_pairs": 0,
      "flip_rate": 0.41509433962264153,
      "probability_drift": 0.12496415094339623,
      "attack_success_rate": 0.5094339622641509
    }
  },
  "warnings": [],
  "license": "CC0-1.0",
  "n": 200,
  "n_valid": 200,
  "schema_validity": 1.0,
  "accuracy": 0.545,
  "accuracy_ci95": [
    0.475,
    0.615
  ],
  "macro_f1": 0.5077755240027045,
  "macro_f1_ci95": [
    0.4384785319627089,
    0.581729243106952
  ],
  "ece": 0.129945,
  "ece_ci95": [
    0.06489222500000003,
    0.198451975
  ],
  "brier": 0.4981771118,
  "brier_ci95": [
    0.4505436749075,
    0.5437308911525001
  ],
  "brier_max": 2.0,
  "score_mae": null,
  "s1_score": 71.1054928196172,
  "s1_score_ci95": [
    67.08669799313768,
    75.28625091514493
  ],
  "components": {
    "accuracy": 0.545,
    "macro_f1": 0.5077755240027045,
    "calibration": 0.870055,
    "brier_quality": 0.7509114441,
    "quality": 0.6381647222209466,
    "schema_validity": 1.0,
    "robustness": 0.7622641509433963,
    "robustness_measured": true
  },
  "composite_id": "s1-v1",
  "reliability": [
    {
      "bin": 0,
      "low": 0.0,
      "high": 0.1,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 1,
      "low": 0.1,
      "high": 0.2,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 2,
      "low": 0.2,
      "high": 0.3,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 3,
      "low": 0.3,
      "high": 0.4,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 4,
      "low": 0.4,
      "high": 0.5,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 5,
      "low": 0.5,
      "high": 0.6,
      "count": 43,
      "confidence": 0.5583837209302327,
      "accuracy": 0.37209302325581395
    },
    {
      "bin": 6,
      "low": 0.6,
      "high": 0.7,
      "count": 87,
      "confidence": 0.650567816091954,
      "accuracy": 0.5287356321839081
    },
    {
      "bin": 7,
      "low": 0.7,
      "high": 0.8,
      "count": 65,
      "confidence": 0.74544,
      "accuracy": 0.6461538461538462
    },
    {
      "bin": 8,
      "low": 0.8,
      "high": 0.9,
      "count": 5,
      "confidence": 0.8149,
      "accuracy": 1.0
    },
    {
      "bin": 9,
      "low": 0.9,
      "high": 1.0,
      "count": 0,
      "confidence": null,
      "accuracy": null
    }
  ],
  "selective": [
    {
      "threshold": 0.5,
      "n": 200,
      "coverage": 1.0,
      "risk": 0.45499999999999996
    },
    {
      "threshold": 0.6,
      "n": 157,
      "coverage": 0.785,
      "risk": 0.40764331210191085
    },
    {
      "threshold": 0.7,
      "n": 70,
      "coverage": 0.35,
      "risk": 0.3285714285714286
    },
    {
      "threshold": 0.8,
      "n": 5,
      "coverage": 0.025,
      "risk": 0.0
    },
    {
      "threshold": 0.9,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    },
    {
      "threshold": 0.95,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    }
  ],
  "risk_coverage": [
    {
      "coverage": 0.1,
      "n": 20,
      "risk": 0.09999999999999998
    },
    {
      "coverage": 0.2,
      "n": 40,
      "risk": 0.25
    },
    {
      "coverage": 0.3,
      "n": 60,
      "risk": 0.31666666666666665
    },
    {
      "coverage": 0.4,
      "n": 80,
      "risk": 0.30000000000000004
    },
    {
      "coverage": 0.5,
      "n": 100,
      "risk": 0.32999999999999996
    },
    {
      "coverage": 0.6,
      "n": 120,
      "risk": 0.3666666666666667
    },
    {
      "coverage": 0.7,
      "n": 140,
      "risk": 0.40714285714285714
    },
    {
      "coverage": 0.8,
      "n": 160,
      "risk": 0.4125
    },
    {
      "coverage": 0.9,
      "n": 180,
      "risk": 0.4388888888888889
    },
    {
      "coverage": 1.0,
      "n": 200,
      "risk": 0.45499999999999996
    }
  ],
  "bootstrap": {
    "n": 1000,
    "seed": 20261006,
    "interval": "percentile 2.5/97.5"
  },
  "notes": {
    "s1_score_interval": "The interval resamples test items and keeps the robustness term fixed at its point estimate. It is not a bootstrap of the probe subset."
  }
}