{
  "checkpoint": "artifacts/partial-shard-warmstart-lr3e4-5ep/best.pt",
  "checkpoint_sha256": "e919cdbdc89046bd82fccf52a460406643e917b06d42a966d0e0104b2974f6c1",
  "data_revision": "e564e2ac567f774d1880aa1db6ce97afb8c519b7",
  "data_scope": "preview-only: 3265 rows from train-00010-of-00083; official test remains sealed",
  "dataset_revision": null,
  "development_only": true,
  "evaluation_source": {
    "bytes": 2791055,
    "dataset_id": null,
    "kind": "file",
    "path": "data/processed/partial-iid-splits.jsonl",
    "revision": "e564e2ac567f774d1880aa1db6ce97afb8c519b7",
    "sha256": "cae0765d18467186c0e890c35fd94c0fcc70ae3483b5a1ce31814dd8ffa37b50"
  },
  "freeze_manifest_sha256": null,
  "grouped_bootstrap_95ci": {
    "false_positive_rate": {
      "estimate": 0.018072289156626505,
      "lower": 0.0,
      "upper": 0.041963307776560786
    },
    "recall": {
      "estimate": 0.0625,
      "lower": 0.028232028053769738,
      "upper": 0.1
    }
  },
  "metrics": {
    "accuracy": 0.5306748466257669,
    "average_precision": 0.6880689959324726,
    "balanced_accuracy": 0.5222138554216867,
    "brier_score": 0.2133037066638398,
    "count": 326,
    "expected_calibration_error": 0.1055241439925747,
    "f1": 0.11560693641618497,
    "false_negative_rate": 0.9375,
    "false_positive_rate": 0.018072289156626505,
    "fn": 150,
    "fp": 3,
    "log_loss": 0.6135001064370812,
    "macro_f1": 0.39809574378220525,
    "negative_count": 166,
    "positive_count": 160,
    "precision": 0.7692307692307693,
    "recall": 0.0625,
    "roc_auc": 0.7389683734939759,
    "specificity": 0.9819277108433735,
    "threshold": 0.741000771522522,
    "tn": 163,
    "tp": 10
  },
  "official_test": false,
  "operating_points": {
    "0.01": {
      "accuracy": 0.5214723926380368,
      "average_precision": 0.6880689959324726,
      "balanced_accuracy": 0.5126129518072289,
      "brier_score": 0.2133037066638398,
      "count": 326,
      "expected_calibration_error": 0.1055241439925747,
      "f1": 0.060240963855421686,
      "false_negative_rate": 0.96875,
      "false_positive_rate": 0.006024096385542169,
      "fn": 155,
      "fp": 1,
      "log_loss": 0.6135001064370812,
      "macro_f1": 0.369626654767217,
      "max_false_positive_rate": 0.01,
      "negative_count": 166,
      "positive_count": 160,
      "precision": 0.8333333333333334,
      "recall": 0.03125,
      "roc_auc": 0.7389683734939759,
      "specificity": 0.9939759036144579,
      "threshold": 0.7609416246414185,
      "tn": 165,
      "tp": 5
    },
    "0.02": {
      "accuracy": 0.5306748466257669,
      "average_precision": 0.6880689959324726,
      "balanced_accuracy": 0.5222138554216867,
      "brier_score": 0.2133037066638398,
      "count": 326,
      "expected_calibration_error": 0.1055241439925747,
      "f1": 0.11560693641618497,
      "false_negative_rate": 0.9375,
      "false_positive_rate": 0.018072289156626505,
      "fn": 150,
      "fp": 3,
      "log_loss": 0.6135001064370812,
      "macro_f1": 0.39809574378220525,
      "max_false_positive_rate": 0.02,
      "negative_count": 166,
      "positive_count": 160,
      "precision": 0.7692307692307693,
      "recall": 0.0625,
      "roc_auc": 0.7389683734939759,
      "specificity": 0.9819277108433735,
      "threshold": 0.741000771522522,
      "tn": 163,
      "tp": 10
    },
    "0.05": {
      "accuracy": 0.5245398773006135,
      "average_precision": 0.6880689959324726,
      "balanced_accuracy": 0.5165286144578313,
      "brier_score": 0.2133037066638398,
      "count": 326,
      "expected_calibration_error": 0.1055241439925747,
      "f1": 0.143646408839779,
      "false_negative_rate": 0.91875,
      "false_positive_rate": 0.04819277108433735,
      "fn": 147,
      "fp": 8,
      "log_loss": 0.6135001064370812,
      "macro_f1": 0.4072796800037536,
      "max_false_positive_rate": 0.05,
      "negative_count": 166,
      "positive_count": 160,
      "precision": 0.6190476190476191,
      "recall": 0.08125,
      "roc_auc": 0.7389683734939759,
      "specificity": 0.9518072289156626,
      "threshold": 0.7245736718177795,
      "tn": 158,
      "tp": 13
    }
  },
  "operating_points_note": "Label-dependent threshold sweep for curve analysis only. The deployed result is `metrics` at the checkpoint's frozen validation-selected threshold.",
  "operational": {
    "accuracy": 0.5306748466257669,
    "audio_hours": 0.7063338715277777,
    "average_precision": 0.6880689959324726,
    "balanced_accuracy": 0.5222138554216867,
    "brier_score": 0.2133037066638398,
    "count": 326,
    "expected_calibration_error": 0.1055241439925747,
    "f1": 0.11560693641618497,
    "false_interruptions": 3,
    "false_interruptions_per_hour": 4.247283219635348,
    "false_negative_rate": 0.9375,
    "false_positive_rate": 0.018072289156626505,
    "fn": 150,
    "fp": 3,
    "log_loss": 0.6135001064370812,
    "macro_f1": 0.39809574378220525,
    "negative_count": 166,
    "positive_count": 160,
    "precision": 0.7692307692307693,
    "recall": 0.0625,
    "roc_auc": 0.7389683734939759,
    "scope": "independent labeled clips; per-hour rate is a clip-normalized proxy, not an online conversation measurement",
    "sequence_metrics_available": false,
    "sequence_metrics_unavailable_reason": "source records contain no genuine turn_id or conversation_id",
    "specificity": 0.9819277108433735,
    "threshold": 0.741000771522522,
    "tn": 163,
    "tp": 10
  },
  "reliability": [
    {
      "count": 8,
      "lower": 0.0,
      "mean_probability": 0.08739477209746838,
      "observed_rate": 0.125,
      "upper": 0.1
    },
    {
      "count": 57,
      "lower": 0.1,
      "mean_probability": 0.14298640361480547,
      "observed_rate": 0.10526315789473684,
      "upper": 0.2
    },
    {
      "count": 56,
      "lower": 0.2,
      "mean_probability": 0.24947392541382993,
      "observed_rate": 0.48214285714285715,
      "upper": 0.3
    },
    {
      "count": 40,
      "lower": 0.3,
      "mean_probability": 0.35491041019558905,
      "observed_rate": 0.35,
      "upper": 0.4
    },
    {
      "count": 43,
      "lower": 0.4,
      "mean_probability": 0.45311658812123673,
      "observed_rate": 0.5813953488372093,
      "upper": 0.5
    },
    {
      "count": 39,
      "lower": 0.5,
      "mean_probability": 0.5574724888190244,
      "observed_rate": 0.6153846153846154,
      "upper": 0.6
    },
    {
      "count": 53,
      "lower": 0.6,
      "mean_probability": 0.6452814441806866,
      "observed_rate": 0.8113207547169812,
      "upper": 0.7
    },
    {
      "count": 30,
      "lower": 0.7,
      "mean_probability": 0.7383094807465871,
      "observed_rate": 0.6666666666666666,
      "upper": 0.8
    },
    {
      "count": 0,
      "lower": 0.8,
      "mean_probability": 0.0,
      "observed_rate": 0.0,
      "upper": 0.9
    },
    {
      "count": 0,
      "lower": 0.9,
      "mean_probability": 0.0,
      "observed_rate": 0.0,
      "upper": 1.0
    }
  ],
  "slices": {
    "dataset": {
      "chirp3_1": {
        "accuracy": 0.5480225988700564,
        "average_precision": 0.6646168482728126,
        "balanced_accuracy": 0.5249615975422427,
        "brier_score": 0.2094916437112573,
        "count": 177,
        "expected_calibration_error": 0.10510466656105666,
        "f1": 0.13043478260869565,
        "false_negative_rate": 0.9285714285714286,
        "false_positive_rate": 0.021505376344086023,
        "fn": 78,
        "fp": 2,
        "log_loss": 0.6028296526829398,
        "macro_f1": 0.412545635579157,
        "negative_count": 93,
        "positive_count": 84,
        "precision": 0.75,
        "recall": 0.07142857142857142,
        "roc_auc": 0.7382232462877624,
        "specificity": 0.978494623655914,
        "threshold": 0.741000771522522,
        "tn": 91,
        "tp": 6
      },
      "chirp3_2": {
        "accuracy": 0.5764705882352941,
        "average_precision": 0.6926460598775176,
        "balanced_accuracy": 0.5166103603603603,
        "brier_score": 0.19349550820379868,
        "count": 85,
        "expected_calibration_error": 0.12449237152057539,
        "f1": 0.1,
        "false_negative_rate": 0.9459459459459459,
        "false_positive_rate": 0.020833333333333332,
        "fn": 35,
        "fp": 1,
        "log_loss": 0.5705930106009551,
        "macro_f1": 0.4115384615384615,
        "negative_count": 48,
        "positive_count": 37,
        "precision": 0.6666666666666666,
        "recall": 0.05405405405405406,
        "roc_auc": 0.7792792792792793,
        "specificity": 0.9791666666666666,
        "threshold": 0.741000771522522,
        "tn": 47,
        "tp": 2
      },
      "liva_1": {
        "accuracy": 0.3611111111111111,
        "average_precision": 0.7967089154342165,
        "balanced_accuracy": 0.5208333333333334,
        "brier_score": 0.28008335716505756,
        "count": 36,
        "expected_calibration_error": 0.26551183147562873,
        "f1": 0.08,
        "false_negative_rate": 0.9583333333333334,
        "false_positive_rate": 0.0,
        "fn": 23,
        "fp": 0,
        "log_loss": 0.7668865136017733,
        "macro_f1": 0.2953191489361702,
        "negative_count": 12,
        "positive_count": 24,
        "precision": 1.0,
        "recall": 0.041666666666666664,
        "roc_auc": 0.6354166666666666,
        "specificity": 1.0,
        "threshold": 0.741000771522522,
        "tn": 12,
        "tp": 1
      }
    },
    "duration_bin": {
      "2-4s": {
        "accuracy": 0.5897435897435898,
        "average_precision": 0.7161800176088101,
        "balanced_accuracy": 0.5986842105263157,
        "brier_score": 0.20126385365673957,
        "count": 39,
        "expected_calibration_error": 0.2071125757617828,
        "f1": 0.38461538461538464,
        "false_negative_rate": 0.75,
        "false_positive_rate": 0.05263157894736842,
        "fn": 15,
        "fp": 1,
        "log_loss": 0.5851101759971097,
        "macro_f1": 0.5384615384615384,
        "negative_count": 19,
        "positive_count": 20,
        "precision": 0.8333333333333334,
        "recall": 0.25,
        "roc_auc": 0.718421052631579,
        "specificity": 0.9473684210526315,
        "threshold": 0.741000771522522,
        "tn": 18,
        "tp": 5
      },
      "4-8s": {
        "accuracy": 0.5337837837837838,
        "average_precision": 0.6484258629260561,
        "balanced_accuracy": 0.5078754578754578,
        "brier_score": 0.22099832665123215,
        "count": 148,
        "expected_calibration_error": 0.11727086781851345,
        "f1": 0.0547945205479452,
        "false_negative_rate": 0.9714285714285714,
        "false_positive_rate": 0.01282051282051282,
        "fn": 68,
        "fp": 1,
        "log_loss": 0.6313563076194476,
        "macro_f1": 0.3726887400946004,
        "negative_count": 78,
        "positive_count": 70,
        "precision": 0.6666666666666666,
        "recall": 0.02857142857142857,
        "roc_auc": 0.7208791208791209,
        "specificity": 0.9871794871794872,
        "threshold": 0.741000771522522,
        "tn": 77,
        "tp": 2
      },
      ">=8s": {
        "accuracy": 0.5147058823529411,
        "average_precision": 0.7332799215478968,
        "balanced_accuracy": 0.5147058823529412,
        "brier_score": 0.21026609608377042,
        "count": 136,
        "expected_calibration_error": 0.10151389422004713,
        "f1": 0.08333333333333333,
        "false_negative_rate": 0.9558823529411765,
        "false_positive_rate": 0.014705882352941176,
        "fn": 65,
        "fp": 1,
        "log_loss": 0.6061258970823683,
        "macro_f1": 0.3766666666666667,
        "negative_count": 68,
        "positive_count": 68,
        "precision": 0.75,
        "recall": 0.04411764705882353,
        "roc_auc": 0.7588667820069204,
        "specificity": 0.9852941176470589,
        "threshold": 0.741000771522522,
        "tn": 67,
        "tp": 3
      }
    },
    "filler_type": {
      "endfiller": {
        "accuracy": 1.0,
        "average_precision": null,
        "balanced_accuracy": 1.0,
        "brier_score": 0.11988233328937586,
        "count": 64,
        "expected_calibration_error": 0.2932384021114558,
        "f1": null,
        "false_negative_rate": null,
        "false_positive_rate": 0.0,
        "fn": 0,
        "fp": 0,
        "log_loss": 0.3894655147868388,
        "macro_f1": 1.0,
        "negative_count": 64,
        "positive_count": 0,
        "precision": null,
        "recall": null,
        "roc_auc": null,
        "specificity": 1.0,
        "threshold": 0.741000771522522,
        "tn": 64,
        "tp": 0
      },
      "midfiller": {
        "accuracy": 0.5522388059701493,
        "average_precision": 0.6342371202899987,
        "balanced_accuracy": 0.5255980326402862,
        "brier_score": 0.2159177382893442,
        "count": 134,
        "expected_calibration_error": 0.12705712759895113,
        "f1": 0.14285714285714285,
        "false_negative_rate": 0.9206349206349206,
        "false_positive_rate": 0.028169014084507043,
        "fn": 58,
        "fp": 2,
        "log_loss": 0.6196895811384678,
        "macro_f1": 0.4199134199134199,
        "negative_count": 71,
        "positive_count": 63,
        "precision": 0.7142857142857143,
        "recall": 0.07936507936507936,
        "roc_auc": 0.7165213503241672,
        "specificity": 0.971830985915493,
        "threshold": 0.741000771522522,
        "tn": 69,
        "tp": 5
      },
      "no_filler": {
        "accuracy": 0.13846153846153847,
        "average_precision": 0.9282220984617144,
        "balanced_accuracy": 0.4544334975369458,
        "brier_score": 0.2636098061789016,
        "count": 65,
        "expected_calibration_error": 0.39574201336273784,
        "f1": 0.0967741935483871,
        "false_negative_rate": 0.9482758620689655,
        "false_positive_rate": 0.14285714285714285,
        "fn": 55,
        "fp": 1,
        "log_loss": 0.7362453768676601,
        "macro_f1": 0.1366223908918406,
        "negative_count": 7,
        "positive_count": 58,
        "precision": 0.75,
        "recall": 0.05172413793103448,
        "roc_auc": 0.6798029556650246,
        "specificity": 0.8571428571428571,
        "threshold": 0.741000771522522,
        "tn": 6,
        "tp": 3
      },
      "unknown": {
        "accuracy": 0.4126984126984127,
        "average_precision": 0.799471612772439,
        "balanced_accuracy": 0.5256410256410257,
        "brier_score": 0.25074483665858716,
        "count": 63,
        "expected_calibration_error": 0.19315069298895576,
        "f1": 0.0975609756097561,
        "false_negative_rate": 0.9487179487179487,
        "false_positive_rate": 0.0,
        "fn": 37,
        "fp": 0,
        "log_loss": 0.7012839425901298,
        "macro_f1": 0.3311334289813486,
        "negative_count": 24,
        "positive_count": 39,
        "precision": 1.0,
        "recall": 0.05128205128205128,
        "roc_auc": 0.6720085470085471,
        "specificity": 1.0,
        "threshold": 0.741000771522522,
        "tn": 24,
        "tp": 2
      }
    },
    "language": {
      "eng": {
        "accuracy": 0.4342105263157895,
        "average_precision": 0.838347854611721,
        "balanced_accuracy": 0.5222222222222223,
        "brier_score": 0.2267736908018496,
        "count": 76,
        "expected_calibration_error": 0.17203282682519208,
        "f1": 0.0851063829787234,
        "false_negative_rate": 0.9555555555555556,
        "false_positive_rate": 0.0,
        "fn": 43,
        "fp": 0,
        "log_loss": 0.6475126882660074,
        "macro_f1": 0.33779128672745695,
        "negative_count": 31,
        "positive_count": 45,
        "precision": 1.0,
        "recall": 0.044444444444444446,
        "roc_auc": 0.7483870967741936,
        "specificity": 1.0,
        "threshold": 0.741000771522522,
        "tn": 31,
        "tp": 2
      }
    },
    "synthetic": {
      "False": {
        "accuracy": 0.4,
        "average_precision": 0.8139063380698389,
        "balanced_accuracy": 0.5285714285714286,
        "brier_score": 0.2517751278920657,
        "count": 55,
        "expected_calibration_error": 0.20835820707407862,
        "f1": 0.10810810810810811,
        "false_negative_rate": 0.9428571428571428,
        "false_positive_rate": 0.0,
        "fn": 33,
        "fp": 0,
        "log_loss": 0.7044581255390688,
        "macro_f1": 0.32802665679378007,
        "negative_count": 20,
        "positive_count": 35,
        "precision": 1.0,
        "recall": 0.05714285714285714,
        "roc_auc": 0.6785714285714286,
        "specificity": 1.0,
        "threshold": 0.741000771522522,
        "tn": 20,
        "tp": 2
      },
      "True": {
        "accuracy": 0.5571955719557196,
        "average_precision": 0.6669351194253922,
        "balanced_accuracy": 0.5217260273972603,
        "brier_score": 0.20549585364704118,
        "count": 271,
        "expected_calibration_error": 0.09752494687853705,
        "f1": 0.11764705882352941,
        "false_negative_rate": 0.936,
        "false_positive_rate": 0.02054794520547945,
        "fn": 117,
        "fp": 3,
        "log_loss": 0.595039991859187,
        "macro_f1": 0.41104027818023764,
        "negative_count": 146,
        "positive_count": 125,
        "precision": 0.7272727272727273,
        "recall": 0.064,
        "roc_auc": 0.7485479452054794,
        "specificity": 0.9794520547945206,
        "threshold": 0.741000771522522,
        "tn": 143,
        "tp": 8
      }
    }
  },
  "source_split": "validation",
  "split": "validation",
  "threshold": 0.741000771522522,
  "training_status": "preview-only",
  "worst_groups": {
    "highest_false_positive_rate": {
      "count": 65,
      "dimension": "filler_type",
      "false_positive_rate": 0.14285714285714285,
      "value": "no_filler"
    },
    "lowest_recall": {
      "count": 148,
      "dimension": "duration_bin",
      "recall": 0.02857142857142857,
      "value": "4-8s"
    }
  }
}