{
  "stage": 3,
  "model_path": "models/stage3-bert-tiny-ner",
  "records": 40,
  "patterns": 10,
  "overall": {
    "true_positive": 50.0,
    "gold_spans": 98.0,
    "predicted_spans": 111.0,
    "precision": 0.45045045045045046,
    "recall": 0.5102040816326531,
    "f1": 0.4784688995215311
  },
  "error_counts": {
    "boundary": 18,
    "false_positive": 22,
    "missed_entity": 9,
    "wrong_label": 21
  },
  "label_confusions": {
    "AMOUNT->PRODUCT": 2,
    "JOB->ORGANIZATION": 2,
    "LOCATION->ORGANIZATION": 2,
    "ORGANIZATION->PRODUCT": 1,
    "ORGANIZATION->WORKOFART": 1,
    "PERSON->WORKOFART": 1,
    "PRODUCT->ORGANIZATION": 5,
    "PRODUCT->PERSON": 1,
    "PRODUCT->WORKOFART": 1,
    "TIMEDATE->AMOUNT": 1,
    "TIMEDATE->ORGANIZATION": 1,
    "WORKOFART->ORGANIZATION": 2,
    "WORKOFART->PERSON": 1
  },
  "by_pattern": {
    "amount_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 4.0,
        "gold_spans": 8.0,
        "predicted_spans": 9.0,
        "precision": 0.4444444444444444,
        "recall": 0.5,
        "f1": 0.47058823529411764
      },
      "error_counts": {
        "boundary": 4,
        "false_positive": 1
      },
      "records_with_errors": 2
    },
    "company_product": {
      "records": 4,
      "metrics": {
        "true_positive": 1.0,
        "gold_spans": 10.0,
        "predicted_spans": 9.0,
        "precision": 0.1111111111111111,
        "recall": 0.1,
        "f1": 0.10526315789473685
      },
      "error_counts": {
        "boundary": 2,
        "false_positive": 2,
        "missed_entity": 3,
        "wrong_label": 4
      },
      "records_with_errors": 4
    },
    "institution_context": {
      "records": 4,
      "metrics": {
        "true_positive": 5.0,
        "gold_spans": 9.0,
        "predicted_spans": 12.0,
        "precision": 0.4166666666666667,
        "recall": 0.5555555555555556,
        "f1": 0.4761904761904762
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 4,
        "missed_entity": 1,
        "wrong_label": 2
      },
      "records_with_errors": 4
    },
    "job_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 6.0,
        "gold_spans": 13.0,
        "predicted_spans": 15.0,
        "precision": 0.4,
        "recall": 0.46153846153846156,
        "f1": 0.42857142857142855
      },
      "error_counts": {
        "boundary": 3,
        "false_positive": 3,
        "missed_entity": 1,
        "wrong_label": 3
      },
      "records_with_errors": 4
    },
    "namedness_repetition": {
      "records": 4,
      "metrics": {
        "true_positive": 7.0,
        "gold_spans": 9.0,
        "predicted_spans": 10.0,
        "precision": 0.7,
        "recall": 0.7777777777777778,
        "f1": 0.7368421052631577
      },
      "error_counts": {
        "false_positive": 1,
        "wrong_label": 2
      },
      "records_with_errors": 3
    },
    "person_alias_handle": {
      "records": 4,
      "metrics": {
        "true_positive": 10.0,
        "gold_spans": 12.0,
        "predicted_spans": 12.0,
        "precision": 0.8333333333333334,
        "recall": 0.8333333333333334,
        "f1": 0.8333333333333334
      },
      "error_counts": {
        "false_positive": 1,
        "missed_entity": 1,
        "wrong_label": 1
      },
      "records_with_errors": 2
    },
    "product_work": {
      "records": 4,
      "metrics": {
        "true_positive": 2.0,
        "gold_spans": 9.0,
        "predicted_spans": 12.0,
        "precision": 0.16666666666666666,
        "recall": 0.2222222222222222,
        "f1": 0.1904761904761905
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 4,
        "missed_entity": 1,
        "wrong_label": 5
      },
      "records_with_errors": 4
    },
    "software_namedness": {
      "records": 4,
      "metrics": {
        "true_positive": 4.0,
        "gold_spans": 10.0,
        "predicted_spans": 11.0,
        "precision": 0.36363636363636365,
        "recall": 0.4,
        "f1": 0.380952380952381
      },
      "error_counts": {
        "boundary": 2,
        "false_positive": 2,
        "missed_entity": 1,
        "wrong_label": 3
      },
      "records_with_errors": 3
    },
    "timedate_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 8.0,
        "gold_spans": 10.0,
        "predicted_spans": 13.0,
        "precision": 0.6153846153846154,
        "recall": 0.8,
        "f1": 0.6956521739130435
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 3,
        "wrong_label": 1
      },
      "records_with_errors": 2
    },
    "work_title_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 3.0,
        "gold_spans": 8.0,
        "predicted_spans": 8.0,
        "precision": 0.375,
        "recall": 0.375,
        "f1": 0.375
      },
      "error_counts": {
        "boundary": 4,
        "false_positive": 1,
        "missed_entity": 1
      },
      "records_with_errors": 3
    }
  },
  "method": "Forty fresh, manually gold-labeled policy probes; exact character-span comparison; errors classified by deterministic overlap rules and then manually reviewed."
}