{
  "benchmark_id": "historical-segmentation-refiner-replay-50-20260903",
  "experiment_date": "2026-09-03",
  "visual_export_date": "2026-09-05",
  "scope": "Three selected illustrative cases from the reused 50-image development corpus. They do not estimate typical production quality.",
  "selection_method": "Largest boundary-F1 improvement, a change close to the full-corpus median boundary-F1 delta, and a clear boundary-F1 regression; each has verified public photo attribution.",
  "raw_mask_origin": "Decoded directly from the hash-verified frozen parent-mask response.",
  "test_mask_origin": "Deterministic offline reconstruction using the exact historical Test postprocessor, source revision 5ceddcf. Original replay did not save standalone Test mask bitmaps. Every reconstructed raw/Test metric matches its recorded result within 1e-12 (observed maximum error 0). No model inference was run.",
  "test_refiner_source_sha256": "d4aa69aad7a7be1b24e8614bf7e76f04e08a7e49eb660add055cacea9c56697d",
  "summary_sha256": "804ae3e53205851a93a3b7ed1f8b0c86667c13ee61f6df075f909f14c865bea6",
  "fixture_sha256": "ac669f9edeb97f87468fcb10809bc1654b43e17b7692408802980ec6284e1bfd",
  "metrics_source_sha256": "fa2e692709112ddb0dda51a5f53e79fdd145456e0b5bced8512b996b3b28bddf",
  "runtime": {
    "python": "3.12.14",
    "numpy": "2.1.3",
    "opencv": "4.10.0"
  },
  "svg_method": "Each SVG path is a union of one-pixel-high horizontal runs of the exact binary foreground or the metric-defined four-connected boundary. Reconstructing all runs reproduces the input grid exactly. No contour smoothing, approximation, crop or padding. Native-sized viewBox matches the untouched photo.",
  "overlay_colors": {
    "raw": "#ffb84d",
    "test": "#24e3c2",
    "reference": "#ffffff"
  },
  "overlay_display": "Keep photos unmodified and layer SVGs in HTML/CSS at the same aspect ratio. Apply opacity to full-mask SVGs if desired; outline SVGs represent exact one-native-pixel boundaries. Use actual image dimension attributes, no object-fit:cover cropping.",
  "cases": [
    {
      "case_id": "coco-surface-000000113051-175",
      "coco_image_id": 113051,
      "target": "wall-stone",
      "track": "surface",
      "selection_role": "clear_improvement",
      "suggested_heading": "A large change on a stone wall",
      "asset_directory": "coco-113051",
      "width": 640,
      "height": 480,
      "photo_source_url": "https://www.flickr.com/photos/celesteh/5946397184/",
      "photo_title": "Parked Boat",
      "photo_author": "Charles Hutchins",
      "photo_author_url": "https://www.flickr.com/photos/celesteh",
      "photo_license": "CC BY 2.0",
      "photo_license_url": "https://creativecommons.org/licenses/by/2.0/",
      "photo_transformations": "None. Byte-for-byte copy of the cached COCO JPEG used by the experiment. No crop, resize, recompression, painting or generated pixels.",
      "photo_attribution": "Photo: [Parked Boat](https://www.flickr.com/photos/celesteh/5946397184/) by [Charles Hutchins](https://www.flickr.com/photos/celesteh), [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). Source image supplied through COCO. Separate benchmark selection overlays added by Rendero.",
      "reference_attribution": "COCO-Stuff authors, CC BY 4.0",
      "reference_license_url": "https://creativecommons.org/licenses/by/4.0/",
      "reference_source_url": "https://github.com/nightrome/cocostuff",
      "reference_verification": "Decoded fixture reference matches the original cached COCO-Stuff label map pixel-for-pixel.",
      "reference_annotation_source_sha256": "d3c5d65340d7d580ea0698a81e31b73a52d94825a6c7d7fe8490fac9eb77e457",
      "rgb_hash_matches_record": true,
      "response_hash_matches_record": true,
      "recorded_rgb_sha256": "7c85fa00820e7dc32e718237da751f718b40799a1087621175f4a509e8aed306",
      "recorded_response_sha256": "00c7dffd7fdcea2899d9fed676f8f54add444fb9f71930123cef4a765d03a562",
      "recomputed_metrics": {
        "raw": {
          "iou": 0.7891362903053863,
          "leakage": 0.027424497232149115,
          "missed": 0.19290428563121476,
          "boundary_f1": 0.26208811020301015
        },
        "current": {
          "iou": 0.8326649523823092,
          "leakage": 0.023498454598613317,
          "missed": 0.15030964703145896,
          "boundary_f1": 0.5523011153037963
        }
      },
      "maximum_metric_difference_from_record": 0.0,
      "changed_mask_pixels_raw_to_test": 9492,
      "masks": {
        "raw": {
          "foreground_pixels": 114168,
          "binary_pixels_sha256": "4ffe4b462813a7f7616120b2461c02a7de96c7d8dcb3ae105d7b1b5aadb39131",
          "foreground_bbox_xyxy_exclusive": [
            0,
            0,
            640,
            417
          ]
        },
        "test": {
          "foreground_pixels": 119710,
          "binary_pixels_sha256": "36f1825e304cc47b7023342c77a0898dd49fac806b41849a585703b00908069d",
          "foreground_bbox_xyxy_exclusive": [
            0,
            0,
            640,
            411
          ]
        },
        "reference": {
          "foreground_pixels": 137576,
          "binary_pixels_sha256": "bf5390cf122d64684c3995e4cb7879e9b98558c7513e8e1624ec8f834be3e14c",
          "foreground_bbox_xyxy_exclusive": [
            0,
            0,
            640,
            414
          ]
        }
      },
      "assets": {
        "photo.jpg": {
          "bytes": 148414,
          "sha256": "08e92ec10d2b3b5e1308e6853d50846fb11f718b57d1c4e32e302f8be7aac6a6"
        },
        "raw-mask.svg": {
          "bytes": 60063,
          "sha256": "dc8c084b96400547cb9b38e0743cebb5c0df136eeba3467538d610054dd0c150",
          "one_pixel_high_runs": 3547,
          "exact_reconstruction_verified": true
        },
        "raw-outline.svg": {
          "bytes": 108940,
          "sha256": "a36927d55645842695a80ac15e4bfc68c2fd5bf1d4165fea4ba7bf5f446e728d",
          "one_pixel_high_runs": 6869,
          "exact_reconstruction_verified": true
        },
        "test-mask.svg": {
          "bytes": 17603,
          "sha256": "2384299966ec3c29fab78578f6d3e4de576622a8a46e334f10798711cd44e457",
          "one_pixel_high_runs": 936,
          "exact_reconstruction_verified": true
        },
        "test-outline.svg": {
          "bytes": 29764,
          "sha256": "34b3b49d3b8d089180b95ad84543ee52c27033fecdee25ea35f5df6b3aa46929",
          "one_pixel_high_runs": 1878,
          "exact_reconstruction_verified": true
        },
        "reference-mask.svg": {
          "bytes": 10622,
          "sha256": "608c3dd9385762a2050ca2621656f13735f58663e8ce3e944c043d3608e191a2",
          "one_pixel_high_runs": 550,
          "exact_reconstruction_verified": true
        },
        "reference-outline.svg": {
          "bytes": 17183,
          "sha256": "5e573f91dd8414fc63e2d376ff5d074b1e9f7229d06642b3579a0eae48a4513b",
          "one_pixel_high_runs": 1090,
          "exact_reconstruction_verified": true
        }
      }
    },
    {
      "case_id": "coco-object-000000175364-1123027",
      "coco_image_id": 175364,
      "target": "oven",
      "track": "object",
      "selection_role": "typical_small_change",
      "suggested_heading": "A small change around an oven",
      "asset_directory": "coco-175364",
      "width": 614,
      "height": 461,
      "photo_source_url": "https://www.flickr.com/photos/yhaconwy-yharowen/6267025876/",
      "photo_title": "Kitchen YHA Rowen",
      "photo_author": "Eric Audige-Soutter",
      "photo_author_url": "https://www.flickr.com/photos/yhaconwy-yharowen",
      "photo_license": "CC BY 2.0",
      "photo_license_url": "https://creativecommons.org/licenses/by/2.0/",
      "photo_transformations": "None. Byte-for-byte copy of the cached COCO JPEG used by the experiment. No crop, resize, recompression, painting or generated pixels.",
      "photo_attribution": "Photo: [Kitchen YHA Rowen](https://www.flickr.com/photos/yhaconwy-yharowen/6267025876/) by [Eric Audige-Soutter](https://www.flickr.com/photos/yhaconwy-yharowen), [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). Source image supplied through COCO. Separate benchmark selection overlays added by Rendero.",
      "reference_attribution": "COCO Consortium, CC BY 4.0",
      "reference_license_url": "https://creativecommons.org/licenses/by/4.0/",
      "reference_source_url": "https://cocodataset.org/#termsofuse",
      "reference_verification": "Fixture polygon vertices match the official COCO instance annotation normalized by native width/height; the original repository rasterizer reconstructs the experiment reference.",
      "reference_annotation_source_sha256": "e8c7f7908f1d7278341fae127d0da654f102f11bd7b21d8aeefa635b8c810b6f",
      "rgb_hash_matches_record": true,
      "response_hash_matches_record": true,
      "recorded_rgb_sha256": "f15fc83473af89019e910fa9879e5b452f9a213b078ddf5521ff5d24dff8a23c",
      "recorded_response_sha256": "8b276a226f3a5fd3e34f375523bd2e09a3ab1d98c8568f7a673652f583135636",
      "recomputed_metrics": {
        "raw": {
          "iou": 0.9408840877460402,
          "leakage": 0.043393302338117376,
          "missed": 0.017168570439598477,
          "boundary_f1": 0.8237372328697986
        },
        "current": {
          "iou": 0.9436629064366291,
          "leakage": 0.042098881853697964,
          "missed": 0.015507095880927657,
          "boundary_f1": 0.8395968653081801
        }
      },
      "maximum_metric_difference_from_record": 0.0,
      "changed_mask_pixels_raw_to_test": 169,
      "masks": {
        "raw": {
          "foreground_pixels": 14841,
          "binary_pixels_sha256": "24a395c32fdc94529503728a22edb8eac57c44d2c15dc65a402c9092c6adb9ec",
          "foreground_bbox_xyxy_exclusive": [
            157,
            245,
            262,
            428
          ]
        },
        "test": {
          "foreground_pixels": 14846,
          "binary_pixels_sha256": "ff2e5a3cd7a7de2708878a7537bd9b76f587cd85e3b9ba7fda1a5b055b31b289",
          "foreground_bbox_xyxy_exclusive": [
            157,
            251,
            262,
            427
          ]
        },
        "reference": {
          "foreground_pixels": 14445,
          "binary_pixels_sha256": "ed94243b25ae67be74df3813a94ed797cf7b79e03670266573d00c72f98c7677",
          "foreground_bbox_xyxy_exclusive": [
            159,
            250,
            262,
            429
          ]
        }
      },
      "assets": {
        "photo.jpg": {
          "bytes": 121389,
          "sha256": "badb2570222f6900a65bfc9d39615216c683ce3c5d03afb85c692a3cf08f323c"
        },
        "raw-mask.svg": {
          "bytes": 3615,
          "sha256": "38192a76ab47e2719a13d80cf42fa6d60add6b7a68d28f9d3dedfe867c8d36bb",
          "one_pixel_high_runs": 183,
          "exact_reconstruction_verified": true
        },
        "raw-outline.svg": {
          "bytes": 6000,
          "sha256": "572831d9c9da0663b1403c009ddf3beff2c6d87449c40e375d47ea629a4b23d9",
          "one_pixel_high_runs": 362,
          "exact_reconstruction_verified": true
        },
        "test-mask.svg": {
          "bytes": 3516,
          "sha256": "72a64a2e8fd6090612a3b3403ea3f49b1d0463f79a8f9e31d9331f65f149b369",
          "one_pixel_high_runs": 177,
          "exact_reconstruction_verified": true
        },
        "test-outline.svg": {
          "bytes": 5845,
          "sha256": "b35d29de6f8429b2555219561d4293124a62cbabea80e556f25ae5ed3006d5fa",
          "one_pixel_high_runs": 352,
          "exact_reconstruction_verified": true
        },
        "reference-mask.svg": {
          "bytes": 3571,
          "sha256": "7aa7305cf052208bd5d0b93e0737fe83f6ef3cea823aa56ec43544537fcff011",
          "one_pixel_high_runs": 180,
          "exact_reconstruction_verified": true
        },
        "reference-outline.svg": {
          "bytes": 5924,
          "sha256": "41ad142b724ad495636b6633f2a0abf673ddd3b2839bdc2c3120d74e895d282f",
          "one_pixel_high_runs": 357,
          "exact_reconstruction_verified": true
        }
      }
    },
    {
      "case_id": "coco-object-000000429598-331603",
      "coco_image_id": 429598,
      "target": "refrigerator",
      "track": "object",
      "selection_role": "regression",
      "suggested_heading": "A refrigerator edge gets worse",
      "asset_directory": "coco-429598",
      "width": 640,
      "height": 480,
      "photo_source_url": "https://www.flickr.com/photos/marysalome/6617685485/",
      "photo_title": "Kitchen",
      "photo_author": "marysalome",
      "photo_author_url": "https://www.flickr.com/photos/marysalome",
      "photo_license": "CC BY 2.0",
      "photo_license_url": "https://creativecommons.org/licenses/by/2.0/",
      "photo_transformations": "None. Byte-for-byte copy of the cached COCO JPEG used by the experiment. No crop, resize, recompression, painting or generated pixels.",
      "photo_attribution": "Photo: [Kitchen](https://www.flickr.com/photos/marysalome/6617685485/) by [marysalome](https://www.flickr.com/photos/marysalome), [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). Source image supplied through COCO. Separate benchmark selection overlays added by Rendero.",
      "reference_attribution": "COCO Consortium, CC BY 4.0",
      "reference_license_url": "https://creativecommons.org/licenses/by/4.0/",
      "reference_source_url": "https://cocodataset.org/#termsofuse",
      "reference_verification": "Fixture polygon vertices match the official COCO instance annotation normalized by native width/height; the original repository rasterizer reconstructs the experiment reference.",
      "reference_annotation_source_sha256": "e8c7f7908f1d7278341fae127d0da654f102f11bd7b21d8aeefa635b8c810b6f",
      "rgb_hash_matches_record": true,
      "response_hash_matches_record": true,
      "recorded_rgb_sha256": "478e2e150503808bac96c260efae8048ea34c8c493e80e26a5a4089528849a97",
      "recorded_response_sha256": "1eee3cc2d06d1bb6fcd97995bf806c04f97851f7acdb98a1afe9f5e8e9a0f9df",
      "recomputed_metrics": {
        "raw": {
          "iou": 0.9611152725759468,
          "leakage": 0.032977155490051584,
          "missed": 0.006316043093656421,
          "boundary_f1": 0.6909754727579105
        },
        "current": {
          "iou": 0.9606914318956203,
          "leakage": 0.03439750079353147,
          "missed": 0.0052662375658279695,
          "boundary_f1": 0.6425574375524485
        }
      },
      "maximum_metric_difference_from_record": 0.0,
      "changed_mask_pixels_raw_to_test": 221,
      "masks": {
        "raw": {
          "foreground_pixels": 59708,
          "binary_pixels_sha256": "ae675aed63e0b8e0c61b6522b86bc012924a003f5e4f73e78e2289cf446a3505",
          "foreground_bbox_xyxy_exclusive": [
            0,
            108,
            191,
            480
          ]
        },
        "test": {
          "foreground_pixels": 59859,
          "binary_pixels_sha256": "6095579ca5995cf8edffce932f21fdca0df7505e0fb525fc962871d3bc894fbf",
          "foreground_bbox_xyxy_exclusive": [
            0,
            108,
            192,
            480
          ]
        },
        "reference": {
          "foreground_pixels": 58106,
          "binary_pixels_sha256": "f9b6223815aa0c85d8f5f20a3d554c9a0425444656db3c22c4919300b146cce5",
          "foreground_bbox_xyxy_exclusive": [
            0,
            107,
            192,
            475
          ]
        }
      },
      "assets": {
        "photo.jpg": {
          "bytes": 211819,
          "sha256": "cf692837b22f590cc75ed7eb9bcabd26772c71e924a35506df081b0956755bb2"
        },
        "raw-mask.svg": {
          "bytes": 7164,
          "sha256": "2feb9d95048a8d0bfbdcdb17faad85bfae0323c7e9f957cc8abc77735f1fe184",
          "one_pixel_high_runs": 372,
          "exact_reconstruction_verified": true
        },
        "raw-outline.svg": {
          "bytes": 11631,
          "sha256": "f2dc193fe4edf9b9223dc95dd1cda7cbadc5267504b548c495cb57496ce94ec3",
          "one_pixel_high_runs": 742,
          "exact_reconstruction_verified": true
        },
        "test-mask.svg": {
          "bytes": 7195,
          "sha256": "af43fd170a6c1f8f694ed90888be74fd63958793c36d75626ee5c93a7cadefb0",
          "one_pixel_high_runs": 374,
          "exact_reconstruction_verified": true
        },
        "test-outline.svg": {
          "bytes": 11692,
          "sha256": "b359d6f4ff2275666ffdcfbaa9fc82c59a331191ade253dbcc2e79746a12b559",
          "one_pixel_high_runs": 746,
          "exact_reconstruction_verified": true
        },
        "reference-mask.svg": {
          "bytes": 7129,
          "sha256": "ca6ec8cb77ef4a1ec2c2103d9e92c3e313ce4287d66b98bb33b53c2666fde3b2",
          "one_pixel_high_runs": 369,
          "exact_reconstruction_verified": true
        },
        "reference-outline.svg": {
          "bytes": 11528,
          "sha256": "56e9c666fef064207e6f42cfaf28771b6d0ccd996b3864f0c42b417c35de0f71",
          "one_pixel_high_runs": 733,
          "exact_reconstruction_verified": true
        }
      }
    }
  ]
}