{
  "run_id": "majority__ag_news__20261006T043710Z",
  "backend": "majority",
  "model_version": "s1bench-majority-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": 0.1899859998957254,
  "latency_p50_ms": 0.0027144999421579996,
  "latency_p95_ms": 0.003433449853673664,
  "latency_mean_ms": 0.0028044387525483216,
  "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": 1.0
    }
  },
  "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": 0.25,
  "accuracy_ci95": [
    0.22125,
    0.28
  ],
  "macro_f1": 0.1,
  "macro_f1_ci95": [
    0.09058341862845445,
    0.10937500000000001
  ],
  "ece": 0.0,
  "ece_ci95": [
    0.0,
    0.03375
  ],
  "brier": 0.75,
  "brier_ci95": [
    0.75,
    0.75
  ],
  "brier_max": 2.0,
  "score_mae": null,
  "s1_score": 59.575,
  "s1_score_ci95": [
    58.438668756397135,
    59.9096875
  ],
  "components": {
    "accuracy": 0.25,
    "macro_f1": 0.1,
    "calibration": 1.0,
    "brier_quality": 0.625,
    "quality": 0.4225,
    "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": 800,
      "confidence": 0.25,
      "accuracy": 0.25
    },
    {
      "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": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 6,
      "low": 0.6,
      "high": 0.7,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 7,
      "low": 0.7,
      "high": 0.8,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 8,
      "low": 0.8,
      "high": 0.9,
      "count": 0,
      "confidence": null,
      "accuracy": null
    },
    {
      "bin": 9,
      "low": 0.9,
      "high": 1.0,
      "count": 0,
      "confidence": null,
      "accuracy": null
    }
  ],
  "selective": [
    {
      "threshold": 0.5,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    },
    {
      "threshold": 0.6,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    },
    {
      "threshold": 0.7,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    },
    {
      "threshold": 0.8,
      "n": 0,
      "coverage": 0.0,
      "risk": null
    },
    {
      "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": 0.775
    },
    {
      "coverage": 0.2,
      "n": 160,
      "risk": 0.75625
    },
    {
      "coverage": 0.3,
      "n": 240,
      "risk": 0.775
    },
    {
      "coverage": 0.4,
      "n": 320,
      "risk": 0.765625
    },
    {
      "coverage": 0.5,
      "n": 400,
      "risk": 0.7625
    },
    {
      "coverage": 0.6,
      "n": 480,
      "risk": 0.7583333333333333
    },
    {
      "coverage": 0.7,
      "n": 560,
      "risk": 0.7607142857142857
    },
    {
      "coverage": 0.8,
      "n": 640,
      "risk": 0.7546875
    },
    {
      "coverage": 0.9,
      "n": 720,
      "risk": 0.7458333333333333
    },
    {
      "coverage": 1.0,
      "n": 800,
      "risk": 0.75
    }
  ],
  "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."
  }
}