{
  "benchmark_id": "historical-segmentation-refiner-replay-50-20260903",
  "stages": [
    {
      "stage": "raw",
      "stage_label": "Saved raw parent mask",
      "sample_count": 50,
      "iou": 0.7541538726549355,
      "boundary_f1": 0.4912881800111514,
      "leakage": 0.10828652418633394,
      "missed": 0.20890050680205605
    },
    {
      "stage": "monday",
      "stage_label": "Historical refiner 9317364",
      "sample_count": 50,
      "iou": 0.7596263331358105,
      "boundary_f1": 0.5182535634057701,
      "leakage": 0.11017146671253539,
      "missed": 0.20298500677767634
    },
    {
      "stage": "rectangle_guard",
      "stage_label": "Rectangle guard 91b8f00",
      "sample_count": 50,
      "iou": 0.7580222035907757,
      "boundary_f1": 0.5146728556800763,
      "leakage": 0.11030393531816557,
      "missed": 0.20469155820920115
    },
    {
      "stage": "current",
      "stage_label": "Test refiner 5ceddcf (2026-09-03)",
      "sample_count": 50,
      "iou": 0.7580222035907757,
      "boundary_f1": 0.5146728556800763,
      "leakage": 0.11030393531816557,
      "missed": 0.20469155820920115
    },
    {
      "stage": "candidate",
      "stage_label": "Opt-in line recovery candidate (unpromoted)",
      "sample_count": 50,
      "iou": 0.7580222035907757,
      "boundary_f1": 0.5146728556800763,
      "leakage": 0.11030393531816557,
      "missed": 0.20469155820920115
    }
  ],
  "paired_comparisons": [
    {
      "before_stage": "raw",
      "after_stage": "current",
      "metric": "iou",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 35,
      "declined_cases": 15,
      "unchanged_cases": 0,
      "mean_before": 0.7541538726549355,
      "mean_after": 0.7580222035907757,
      "mean_delta": 0.0038683309358401143,
      "mean_delta_percentage_points": 0.3868330935840114,
      "median_delta": 0.0021356257596432116,
      "minimum_case_delta": -0.003494275127466362,
      "maximum_case_delta": 0.04352866207692285
    },
    {
      "before_stage": "raw",
      "after_stage": "current",
      "metric": "boundary_f1",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 37,
      "declined_cases": 13,
      "unchanged_cases": 0,
      "mean_before": 0.4912881800111514,
      "mean_after": 0.5146728556800763,
      "mean_delta": 0.023384675668924864,
      "mean_delta_percentage_points": 2.3384675668924864,
      "median_delta": 0.01587108514977764,
      "minimum_case_delta": -0.07683608988594566,
      "maximum_case_delta": 0.2902130051007862
    },
    {
      "before_stage": "raw",
      "after_stage": "current",
      "metric": "leakage",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 20,
      "declined_cases": 30,
      "unchanged_cases": 0,
      "mean_before": 0.10828652418633394,
      "mean_after": 0.11030393531816557,
      "mean_delta": 0.0020174111318316105,
      "mean_delta_percentage_points": 0.20174111318316104,
      "median_delta": 0.0004199217912965468,
      "minimum_case_delta": -0.03196261610241588,
      "maximum_case_delta": 0.12992800528239445
    },
    {
      "before_stage": "raw",
      "after_stage": "current",
      "metric": "missed",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 42,
      "declined_cases": 8,
      "unchanged_cases": 0,
      "mean_before": 0.20890050680205605,
      "mean_after": 0.20469155820920115,
      "mean_delta": -0.004208948592854898,
      "mean_delta_percentage_points": -0.42089485928548975,
      "median_delta": -0.002225491031594724,
      "minimum_case_delta": -0.04259463859975579,
      "maximum_case_delta": 0.003500175008750439
    },
    {
      "before_stage": "monday",
      "after_stage": "current",
      "metric": "iou",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 10,
      "unchanged_cases": 40,
      "mean_before": 0.7596263331358105,
      "mean_after": 0.7580222035907757,
      "mean_delta": -0.0016041295450348824,
      "mean_delta_percentage_points": -0.16041295450348825,
      "median_delta": 0.0,
      "minimum_case_delta": -0.020982327560730596,
      "maximum_case_delta": 0.0
    },
    {
      "before_stage": "monday",
      "after_stage": "current",
      "metric": "boundary_f1",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 10,
      "unchanged_cases": 40,
      "mean_before": 0.5182535634057701,
      "mean_after": 0.5146728556800763,
      "mean_delta": -0.0035807077256938434,
      "mean_delta_percentage_points": -0.35807077256938435,
      "median_delta": 0.0,
      "minimum_case_delta": -0.0840312879443309,
      "maximum_case_delta": 0.0
    },
    {
      "before_stage": "monday",
      "after_stage": "current",
      "metric": "leakage",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 10,
      "unchanged_cases": 40,
      "mean_before": 0.11017146671253539,
      "mean_after": 0.11030393531816557,
      "mean_delta": 0.00013246860563018064,
      "mean_delta_percentage_points": 0.013246860563018064,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.0018886082535663484
    },
    {
      "before_stage": "monday",
      "after_stage": "current",
      "metric": "missed",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 10,
      "unchanged_cases": 40,
      "mean_before": 0.20298500677767634,
      "mean_after": 0.20469155820920115,
      "mean_delta": 0.001706551431524797,
      "mean_delta_percentage_points": 0.1706551431524797,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.02192508414909057
    },
    {
      "before_stage": "current",
      "after_stage": "candidate",
      "metric": "iou",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 0,
      "unchanged_cases": 50,
      "mean_before": 0.7580222035907757,
      "mean_after": 0.7580222035907757,
      "mean_delta": 0.0,
      "mean_delta_percentage_points": 0.0,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.0
    },
    {
      "before_stage": "current",
      "after_stage": "candidate",
      "metric": "boundary_f1",
      "better_direction": "higher",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 0,
      "unchanged_cases": 50,
      "mean_before": 0.5146728556800763,
      "mean_after": 0.5146728556800763,
      "mean_delta": 0.0,
      "mean_delta_percentage_points": 0.0,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.0
    },
    {
      "before_stage": "current",
      "after_stage": "candidate",
      "metric": "leakage",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 0,
      "unchanged_cases": 50,
      "mean_before": 0.11030393531816557,
      "mean_after": 0.11030393531816557,
      "mean_delta": 0.0,
      "mean_delta_percentage_points": 0.0,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.0
    },
    {
      "before_stage": "current",
      "after_stage": "candidate",
      "metric": "missed",
      "better_direction": "lower",
      "sample_count": 50,
      "improved_cases": 0,
      "declined_cases": 0,
      "unchanged_cases": 50,
      "mean_before": 0.20469155820920115,
      "mean_after": 0.20469155820920115,
      "mean_delta": 0.0,
      "mean_delta_percentage_points": 0.0,
      "median_delta": 0.0,
      "minimum_case_delta": 0.0,
      "maximum_case_delta": 0.0
    }
  ]
}
