{
  "stage": 4,
  "model_path": "models/stage4-bert-tiny-ner-improved",
  "records": 40,
  "patterns": 10,
  "overall": {
    "true_positive": 62.0,
    "gold_spans": 98.0,
    "predicted_spans": 114.0,
    "precision": 0.543859649122807,
    "recall": 0.6326530612244898,
    "f1": 0.5849056603773586
  },
  "error_counts": {
    "boundary": 18,
    "false_positive": 20,
    "missed_entity": 4,
    "wrong_label": 14
  },
  "label_confusions": {
    "LOCATION->ORGANIZATION": 1,
    "ORGANIZATION->PRODUCT": 1,
    "ORGANIZATION->WORKOFART": 1,
    "PERSON->PRODUCT": 1,
    "PRODUCT->ORGANIZATION": 3,
    "PRODUCT->PERSON": 1,
    "PRODUCT->WORKOFART": 1,
    "TIMEDATE->ORGANIZATION": 1,
    "TIMEDATE->PRODUCT": 1,
    "WORKOFART->LOCATION": 1,
    "WORKOFART->ORGANIZATION": 1,
    "WORKOFART->PERSON": 1
  },
  "by_pattern": {
    "amount_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 8.0,
        "gold_spans": 8.0,
        "predicted_spans": 8.0,
        "precision": 1.0,
        "recall": 1.0,
        "f1": 1.0
      },
      "error_counts": {},
      "records_with_errors": 0
    },
    "company_product": {
      "records": 4,
      "metrics": {
        "true_positive": 4.0,
        "gold_spans": 10.0,
        "predicted_spans": 10.0,
        "precision": 0.4,
        "recall": 0.4,
        "f1": 0.4000000000000001
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 1,
        "missed_entity": 1,
        "wrong_label": 4
      },
      "records_with_errors": 4
    },
    "institution_context": {
      "records": 4,
      "metrics": {
        "true_positive": 7.0,
        "gold_spans": 9.0,
        "predicted_spans": 11.0,
        "precision": 0.6363636363636364,
        "recall": 0.7777777777777778,
        "f1": 0.7000000000000001
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 2,
        "wrong_label": 1
      },
      "records_with_errors": 3
    },
    "job_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 8.0,
        "gold_spans": 13.0,
        "predicted_spans": 16.0,
        "precision": 0.5,
        "recall": 0.6153846153846154,
        "f1": 0.5517241379310345
      },
      "error_counts": {
        "boundary": 5,
        "false_positive": 3
      },
      "records_with_errors": 3
    },
    "namedness_repetition": {
      "records": 4,
      "metrics": {
        "true_positive": 8.0,
        "gold_spans": 9.0,
        "predicted_spans": 9.0,
        "precision": 0.8888888888888888,
        "recall": 0.8888888888888888,
        "f1": 0.8888888888888888
      },
      "error_counts": {
        "wrong_label": 1
      },
      "records_with_errors": 1
    },
    "person_alias_handle": {
      "records": 4,
      "metrics": {
        "true_positive": 9.0,
        "gold_spans": 12.0,
        "predicted_spans": 12.0,
        "precision": 0.75,
        "recall": 0.75,
        "f1": 0.75
      },
      "error_counts": {
        "boundary": 1,
        "wrong_label": 2
      },
      "records_with_errors": 2
    },
    "product_work": {
      "records": 4,
      "metrics": {
        "true_positive": 2.0,
        "gold_spans": 9.0,
        "predicted_spans": 17.0,
        "precision": 0.11764705882352941,
        "recall": 0.2222222222222222,
        "f1": 0.15384615384615383
      },
      "error_counts": {
        "boundary": 2,
        "false_positive": 9,
        "missed_entity": 1,
        "wrong_label": 4
      },
      "records_with_errors": 4
    },
    "software_namedness": {
      "records": 4,
      "metrics": {
        "true_positive": 6.0,
        "gold_spans": 10.0,
        "predicted_spans": 11.0,
        "precision": 0.5454545454545454,
        "recall": 0.6,
        "f1": 0.5714285714285713
      },
      "error_counts": {
        "boundary": 1,
        "false_positive": 2,
        "missed_entity": 1,
        "wrong_label": 2
      },
      "records_with_errors": 3
    },
    "timedate_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 8.0,
        "gold_spans": 10.0,
        "predicted_spans": 12.0,
        "precision": 0.6666666666666666,
        "recall": 0.8,
        "f1": 0.7272727272727272
      },
      "error_counts": {
        "boundary": 2,
        "false_positive": 2
      },
      "records_with_errors": 2
    },
    "work_title_boundary": {
      "records": 4,
      "metrics": {
        "true_positive": 2.0,
        "gold_spans": 8.0,
        "predicted_spans": 8.0,
        "precision": 0.25,
        "recall": 0.25,
        "f1": 0.25
      },
      "error_counts": {
        "boundary": 5,
        "false_positive": 1,
        "missed_entity": 1
      },
      "records_with_errors": 4
    }
  },
  "method": "Forty fresh, manually gold-labeled policy probes; exact character-span comparison; errors classified by deterministic overlap rules and then manually reviewed."
}