{
  "run_id": "random__ag_news__20261006T043912Z",
  "backend": "random",
  "model_version": "s1bench-random-1",
  "suite_id": "ag_news",
  "suite_version": "1.0.0",
  "suite_title": "AG News topic",
  "primitive": "choice",
  "protocol": "1.0.0",
  "split": "test",
  "limited": false,
  "limit": null,
  "prepare_ms": 0.0,
  "latency_p50_ms": 0.01071250005679758,
  "latency_p95_ms": 0.014319650063043804,
  "latency_mean_ms": 0.01195427250110015,
  "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-glibc2.39",
    "python": "3.12.3",
    "machine": "x86_64",
    "processor": "x86_64",
    "cpu_count": 4
  },
  "robustness_probes": {
    "order_reverse": {
      "n": 200,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "order_shuffle": {
      "n": 200,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "paraphrase": {
      "n": 200,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "irrelevant_context": {
      "n": 200,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": null
    },
    "adversarial": {
      "n": 200,
      "invalid_pairs": 0,
      "flip_rate": 0.0,
      "probability_drift": 0.0,
      "attack_success_rate": 0.34
    }
  },
  "warnings": [],
  "license": "License unknown on the Hugging Face card. Research corpus from Zhang, Zhao, and LeCun (2015). Frozen subsample only.",
  "n": 800,
  "n_valid": 800,
  "schema_validity": 1.0,
  "accuracy": 0.23625,
  "accuracy_ci95": [
    0.2075,
    0.26503125
  ],
  "macro_f1": 0.23608436793654586,
  "macro_f1_ci95": [
    0.2059653295181792,
    0.2646120106760326
  ],
  "ece": 0.17907573404200242,
  "ece_ci95": [
    0.1524030610969456,
    0.21247246520610125
  ],
  "brier": 0.8301371482540076,
  "brier_ci95": [
    0.8116619976319601,
    0.8492348167456545
  ],
  "brier_max": 2.0,
  "score_mae": null,
  "s1_score": 59.60042170007929,
  "s1_score_ci95": [
    57.86908598846839,
    61.23998250007345
  ],
  "components": {
    "accuracy": 0.23625,
    "macro_f1": 0.23608436793654586,
    "calibration": 0.8209242659579976,
    "brier_quality": 0.5849314258729962,
    "quality": 0.4228631671439898,
    "schema_validity": 1.0,
    "robustness": 1.0,
    "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": 42,
      "confidence": 0.2845599230775524,
      "accuracy": 0.2857142857142857
    },
    {
      "bin": 3,
      "low": 0.3,
      "high": 0.4,
      "count": 357,
      "confidence": 0.35456005807502383,
      "accuracy": 0.24929971988795518
    },
    {
      "bin": 4,
      "low": 0.4,
      "high": 0.5,
      "count": 271,
      "confidence": 0.4435806751388773,
      "accuracy": 0.22509225092250923
    },
    {
      "bin": 5,
      "low": 0.5,
      "high": 0.6,
      "count": 103,
      "confidence": 0.5362646364901816,
      "accuracy": 0.1941747572815534
    },
    {
      "bin": 6,
      "low": 0.6,
      "high": 0.7,
      "count": 19,
      "confidence": 0.6419987739873687,
      "accuracy": 0.3684210526315789
    },
    {
      "bin": 7,
      "low": 0.7,
      "high": 0.8,
      "count": 7,
      "confidence": 0.7320356372901234,
      "accuracy": 0.0
    },
    {
      "bin": 8,
      "low": 0.8,
      "high": 0.9,
      "count": 1,
      "confidence": 0.866316582160296,
      "accuracy": 0.0
    },
    {
      "bin": 9,
      "low": 0.9,
      "high": 1.0,
      "count": 0,
      "confidence": null,
      "accuracy": null
    }
  ],
  "selective": [
    {
      "threshold": 0.5,
      "n": 130,
      "coverage": 0.1625,
      "risk": 0.7923076923076923
    },
    {
      "threshold": 0.6,
      "n": 27,
      "coverage": 0.03375,
      "risk": 0.7407407407407407
    },
    {
      "threshold": 0.7,
      "n": 8,
      "coverage": 0.01,
      "risk": 1.0
    },
    {
      "threshold": 0.8,
      "n": 1,
      "coverage": 0.00125,
      "risk": 1.0
    },
    {
      "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": 80,
      "risk": 0.7875
    },
    {
      "coverage": 0.2,
      "n": 160,
      "risk": 0.7875
    },
    {
      "coverage": 0.3,
      "n": 240,
      "risk": 0.7833333333333333
    },
    {
      "coverage": 0.4,
      "n": 320,
      "risk": 0.7875
    },
    {
      "coverage": 0.5,
      "n": 400,
      "risk": 0.78
    },
    {
      "coverage": 0.6,
      "n": 480,
      "risk": 0.7666666666666666
    },
    {
      "coverage": 0.7,
      "n": 560,
      "risk": 0.7625
    },
    {
      "coverage": 0.8,
      "n": 640,
      "risk": 0.771875
    },
    {
      "coverage": 0.9,
      "n": 720,
      "risk": 0.7694444444444444
    },
    {
      "coverage": 1.0,
      "n": 800,
      "risk": 0.76375
    }
  ],
  "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."
  }
}