Ink Bench: examples on all 37 windings, researcher takeaway (rank by AP, report the X-ray floor), ink_9um credit, untrack pycache, sturdier leaderboard test

README.md

@@ -26,7 +26,7 @@ PHerc. 139 is a papyrus roll carbonized when Vesuvius buried Herculaneum in 79 C
 - **Shapes**: the 9.362 µm tifxyz meshes of the 37 traced windings (`w023` to `w059`), decimated to 579,450 vertices.
 - **X-ray layer**: the CT intensity on each surface, from the public 9.362 µm surface volumes.
 - **Published ink (2.4 µm), amber**: the Vesuvius Challenge's public ink detection on the 2.4 µm scan. In 3D it is drawn through a local-contrast filter so letter-sized strokes stand out. The close-up view shows the published preview without filtering, downscaled to 2,400 px on its long side and re-saved as JPEG.
-- **Our ink model (9 µm), cyan**: inked by running the public [`scrollprize/ink_9um`](https://huggingface.co/scrollprize/ink_9um) model (seed 42, step 75000) on the 9.362 µm scan. So far we have run the public `scrollprize/ink_9um` model on Hugging Face ZeroGPU over windings `w046` to `w051`. More windings are added as daily free GPU runs allow, so this layer is a work in progress and will grow. The first batch was 37,141 patches in 57 s of measured GPU time. The page reads the covered windings from the manifest, so its text updates as windings are added. The 9.362 µm scan is about four times coarser than the 2.4 µm one, so this layer is noisier. `w047` and `w050` ran with patch overlap 0.5; the other four ran with 0.25 or 0.2, and their 128 px patch grid shows as faint square tiles. Per-run records (checkpoint sha256, villa commit, requested and measured GPU seconds) are in `data/gpu/receipts/`; `data/gpu.json` is the summary the page reads.
+- **Our ink model (9 µm), cyan**: inked by running the public [`scrollprize/ink_9um`](https://huggingface.co/scrollprize/ink_9um) model (seed 42, step 75000) on the 9.362 µm scan. The ink_9um model was trained by the Vesuvius Challenge and run by Chase Hendrick, on Hugging Face ZeroGPU and Modal, over all 37 windings (`w023` to `w059`). The first batch was 37,141 patches in 57 s of measured GPU time. The page reads the covered windings from the manifest, so its text updates as windings are added. The 9.362 µm scan is about four times coarser than the 2.4 µm one, so this layer is noisier. `w047` and `w050` ran with patch overlap 0.5; others ran with 0.25 or 0.2, and their 128 px patch grid shows as faint square tiles. Per-run records (checkpoint sha256, villa commit, requested and measured GPU seconds) are in `data/gpu/receipts/`; `data/gpu.json` is the summary the page reads.
 
 Unrolling lays the windings end to end, outermost first. Each traced winding runs a little past one full turn, so roughly a tenth of the 2.8 m strip repeats surface from its neighbour.
 
@@ -57,3 +57,7 @@ Data: Giorgio Angelotti, Stephen Parsons, Sean Johnson, Elian Rafael Dal Prà, J
 ## Ink Bench
 
 Score your own ink prediction for any winding against the published ink, in the browser (nothing is uploaded): automatic UV alignment, side by side, swipe and difference views, Pearson r, IoU, Dice and a precision and recall curve (agreement with another model's output, not ground truth), letter boxes marked "candidate for expert review", and a one-click reproducible bundle. `tools/inkbench.py` gives identical results in Python and on the command line. See [docs/BENCH.md](docs/BENCH.md) and the open bundle format [docs/BENCH_FORMAT.md](docs/BENCH_FORMAT.md). The Bench code and format are Apache-2.0 (`LICENSE-APACHE`, `NOTICE`, by Chase Hendrick); the data stays CC BY-NC 4.0.
+
+### Researcher takeaway
+
+On the public ink_9um run over all 37 windings (`docs/examples/bench/`), against the published 2.4 µm ink: median AP 0.319 for ink_9um, 0.117 for raw X-ray intensity and 0.089 for uniform noise, and ink_9um beats X-ray intensity on AP on every winding. On Pearson r the medians are 0.422, 0.266 and 0.000, but X-ray intensity outranks ink_9um on `w039`, `w053`, `w054` and `w057`. So: rank ink models by AP (threshold-free and robust to brightness), always report the X-ray floor next to your model, and look at windings where the floor wins before trusting a correlation. These are agreement scores with another model's output, not ground truth.

data/bench/index.json

@@ -17,6 +17,10 @@
     "ct": {
      "file": "data/bench/w023_ct.png",
      "sha256": "9e871edfb98d9c5455e3e8b6dae5bc2127b4b6c0ab2399ea5f80c9b9a38b1d2b"
+    },
+    "gpu": {
+     "file": "data/bench/w023_gpu.png",
+     "sha256": "14bedeb6a276f50d4783caec9774ccc4cf5899ddd67a0d57bf75dc6d04372d0c"
     }
    }
   },
@@ -34,6 +38,10 @@
     "ct": {
      "file": "data/bench/w024_ct.png",
      "sha256": "7f21087fc1b5a1a5812e02d24134004524373408d5fe533939f12003f90c6517"
+    },
+    "gpu": {
+     "file": "data/bench/w024_gpu.png",
+     "sha256": "cc11fb74a30bc3fb48bd652946e3e9124d1a2d90947f613f715da37671a85efd"
     }
    }
   },
@@ -51,6 +59,10 @@
     "ct": {
      "file": "data/bench/w025_ct.png",
      "sha256": "e30f050d29ccc83eaf9b29beafe66f0a47a35ac2a7f30dbd525f462c9ea325fa"
+    },
+    "gpu": {
+     "file": "data/bench/w025_gpu.png",
+     "sha256": "c129d596d1fe215b2a5c671eff5143e938bd4c4d34ac1a6c7b9843d1f263f226"
     }
    }
   },
@@ -68,6 +80,10 @@
     "ct": {
      "file": "data/bench/w026_ct.png",
      "sha256": "d005b08b88e31ad0cc24bfbbff91041e8abad6275d39cbb243dddfe78b46f3f0"
+    },
+    "gpu": {
+     "file": "data/bench/w026_gpu.png",
+     "sha256": "262b1fb5329f95649d5d69da657ac3b111b9ba05c2151859c167efd85c700079"
     }
    }
   },
@@ -85,6 +101,10 @@
     "ct": {
      "file": "data/bench/w027_ct.png",
      "sha256": "bf5b5d9a5d8574c03dba35ac6b502319aa6ebee01032cff01cc5f7d600ea22b6"
+    },
+    "gpu": {
+     "file": "data/bench/w027_gpu.png",
+     "sha256": "89395044a5bf932b7adaeb8522e2d36c87e684f852b6c8ba2e0d30509376e650"
     }
    }
   },
@@ -102,6 +122,10 @@
     "ct": {
      "file": "data/bench/w028_ct.png",
      "sha256": "02e88630f83b2fea27ec4473e8bf04d43a5c771d8b62e8900fd8f56dc4cc3004"
+    },
+    "gpu": {
+     "file": "data/bench/w028_gpu.png",
+     "sha256": "f5b8b736eafc7bd21fd847360d4b9f676d5d94152eb3a5b0b551b5dda437d48e"
     }
    }
   },
@@ -119,6 +143,10 @@
     "ct": {
      "file": "data/bench/w029_ct.png",
      "sha256": "2217cff5ce56140fec1f2f34a055b94ac04a9071d42f20f7d3e9887db23e9025"
+    },
+    "gpu": {
+     "file": "data/bench/w029_gpu.png",
+     "sha256": "2c3c563b8d544fd58025674d718c5aa3393127c710bbe8eb636b699bea6226bb"
     }
    }
   },
@@ -136,6 +164,10 @@
     "ct": {
      "file": "data/bench/w030_ct.png",
      "sha256": "1e739695dddf66e512678c896f56e9fbb5dbb7ac80248db08a5432892f253cd6"
+    },
+    "gpu": {
+     "file": "data/bench/w030_gpu.png",
+     "sha256": "d0bfda06404540de9f895716b7cecbb7b87d8281625bdc77b0c21f0e58bab99b"
     }
    }
   },
@@ -153,6 +185,10 @@
     "ct": {
      "file": "data/bench/w031_ct.png",
      "sha256": "371e1428275d44a00eb5bddf4f87ce93ddb6803cc134537e3304823ee2412be3"
+    },
+    "gpu": {
+     "file": "data/bench/w031_gpu.png",
+     "sha256": "93fc5e9939c05dadc090dd0bb77090fe4016b9d3b45cc5bb0469beaa746a208c"
     }
    }
   },
@@ -170,6 +206,10 @@
     "ct": {
      "file": "data/bench/w032_ct.png",
      "sha256": "83a374f9ad2d93d0fcb30d1fcc6e632830fbd9b6d2bb6aa920c8ae2562674fd1"
+    },
+    "gpu": {
+     "file": "data/bench/w032_gpu.png",
+     "sha256": "ec48088615f9f7a3f80bfe61d130362a50d59d7de3d9bf2565f21f12212efbc5"
     }
    }
   },
@@ -187,6 +227,10 @@
     "ct": {
      "file": "data/bench/w033_ct.png",
      "sha256": "3ccb04488d2c8f5e8eea00f6c690508d5eb49c3c75ba8b1dfada08fb6038139c"
+    },
+    "gpu": {
+     "file": "data/bench/w033_gpu.png",
+     "sha256": "da90168078510f68b84b28617e270f0a33a85d374b9b0db0c565de2365511b1c"
     }
    }
   },
@@ -204,6 +248,10 @@
     "ct": {
      "file": "data/bench/w034_ct.png",
      "sha256": "0b0726cbc9d4fbf6a3851b580b4711606a59c7d221efbefebfe0af350cf74a5d"
+    },
+    "gpu": {
+     "file": "data/bench/w034_gpu.png",
+     "sha256": "015c6a53edb0e9c97c6bacd0f41a38b3ce42af3e1031a7debbdece3771e83dd8"
     }
    }
   },
@@ -221,6 +269,10 @@
     "ct": {
      "file": "data/bench/w035_ct.png",
      "sha256": "65b1c06363cf835d4b008c7155d3e4140c47489c9d1c905ad20baebeb5a18987"
+    },
+    "gpu": {
+     "file": "data/bench/w035_gpu.png",
+     "sha256": "64fe35009ee7787f8d3afa7b41e2a62936fa8a1a00b80376d5013ad7229175da"
     }
    }
   },
@@ -238,6 +290,10 @@
     "ct": {
      "file": "data/bench/w036_ct.png",
      "sha256": "fe43a9e3eb5bed08041cc55a43134b63fb23a85d543f4e89d838350c7de55f0d"
+    },
+    "gpu": {
+     "file": "data/bench/w036_gpu.png",
+     "sha256": "f326573cb16b362511bbbd8e11d51d9cae638c572134f5bf9acd83f983a7799e"
     }
    }
   },
@@ -255,6 +311,10 @@
     "ct": {
      "file": "data/bench/w037_ct.png",
      "sha256": "2656acecedda3f2647058e84185a9d9a5f691272ed0fa99e1c007ab2590a6733"
+    },
+    "gpu": {
+     "file": "data/bench/w037_gpu.png",
+     "sha256": "4b1f3efc88ebc02f44ade6d426371c7ceaa0f45a12d68828cae8a2a0948bcb33"
     }
    }
   },
@@ -272,6 +332,10 @@
     "ct": {
      "file": "data/bench/w038_ct.png",
      "sha256": "3a7dc45d62f9da8ab65a46e3f388813749b058d1e93dfc6bed78d70cde7a76ee"
+    },
+    "gpu": {
+     "file": "data/bench/w038_gpu.png",
+     "sha256": "a9e4b402fa9bc449ef35f0eb3edd3758ad1afafe4ac0178d05cd4245eaa74201"
     }
    }
   },
@@ -289,6 +353,10 @@
     "ct": {
      "file": "data/bench/w039_ct.png",
      "sha256": "927048bd37ec6e3838b6e957fc0dfc4c569f096688b088d48fca0a0dda8c2dbc"
+    },
+    "gpu": {
+     "file": "data/bench/w039_gpu.png",
+     "sha256": "373ddb854a2af7236703c3e4ebb58b148c335230fed2bb3e1498e8136c2c2870"
     }
    }
   },
@@ -306,6 +374,10 @@
     "ct": {
      "file": "data/bench/w040_ct.png",
      "sha256": "df430900b97f8459ea94ad284960bf61051731aca00b48db73ca07264d216e76"
+    },
+    "gpu": {
+     "file": "data/bench/w040_gpu.png",
+     "sha256": "5114fe17084c037b12c0dd883d93e65b287bdabce02e19d22fc1bfa0da0fd518"
     }
    }
   },
@@ -323,6 +395,10 @@
     "ct": {
      "file": "data/bench/w041_ct.png",
      "sha256": "b4e58718a814135c14e2f7d6a2c556c259e9eaac0f72c4eb0abe52b8a288940f"
+    },
+    "gpu": {
+     "file": "data/bench/w041_gpu.png",
+     "sha256": "d57c948b18da6c3c15569b858ad0ee34af03c97bb25bf51d2959e123115127c5"
     }
    }
   },
@@ -340,6 +416,10 @@
     "ct": {
      "file": "data/bench/w042_ct.png",
      "sha256": "0231a5f2b89abba26b3983869727145b140a3d3f0f04b86911c736249d0fb097"
+    },
+    "gpu": {
+     "file": "data/bench/w042_gpu.png",
+     "sha256": "726d4bc78e3256dc7076c9eae37d342f3253a436ee49a0a8993372f6c9746d3b"
     }
    }
   },
@@ -357,6 +437,10 @@
     "ct": {
      "file": "data/bench/w043_ct.png",
      "sha256": "8fe83037cdab7f905bcd55d583950e71956d8729576151cd6e93b9b5610a8a36"
+    },
+    "gpu": {
+     "file": "data/bench/w043_gpu.png",
+     "sha256": "b92f6007d57a05b7c08958c60b5070e70f895a445a12f740f2ec8165d4be42ea"
     }
    }
   },
@@ -374,6 +458,10 @@
     "ct": {
      "file": "data/bench/w044_ct.png",
      "sha256": "b537b3e4505f07bd871dfc6a680471ba88de93c354e0ef8c6d825c009039290c"
+    },
+    "gpu": {
+     "file": "data/bench/w044_gpu.png",
+     "sha256": "b5148db74ef0a5a4c705b7a6928656e36cbfaee0af8f42ed84bedb8df85199ae"
     }
    }
   },
@@ -391,6 +479,10 @@
     "ct": {
      "file": "data/bench/w045_ct.png",
      "sha256": "0d991e625d4abb1ffce31f904a3e47ed012cdfe291f02ceecd5fdf733fc4662b"
+    },
+    "gpu": {
+     "file": "data/bench/w045_gpu.png",
+     "sha256": "fe8cd07604368288da0eb2d4da667ca69d0349b7646814a6db77438d9d0c7bf9"
     }
    }
   },
@@ -534,6 +626,10 @@
     "ct": {
      "file": "data/bench/w052_ct.png",
      "sha256": "29b475c37e05f7a8078e2eadedbbf1863899ca48e39caec2e7fb76d91a508154"
+    },
+    "gpu": {
+     "file": "data/bench/w052_gpu.png",
+     "sha256": "c79565130088f5a8d0dd423ab64bebcf604e871a7a0960632578227e21f72278"
     }
    }
   },
@@ -551,6 +647,10 @@
     "ct": {
      "file": "data/bench/w053_ct.png",
      "sha256": "bb50da6580261cef973e4276ec7936d114d55845c8efac9088ea2c5a91f76e5d"
+    },
+    "gpu": {
+     "file": "data/bench/w053_gpu.png",
+     "sha256": "09c8e76abc3c5f1d4d979709bee06692946fd8b9b1054d262234ab41717ad319"
     }
    }
   },
@@ -568,6 +668,10 @@
     "ct": {
      "file": "data/bench/w054_ct.png",
      "sha256": "a6c5afcbeb31180ebfc9f286802fd07f01ae517a3db502e3972d73b68d6dd0b6"
+    },
+    "gpu": {
+     "file": "data/bench/w054_gpu.png",
+     "sha256": "c3b3cf2331601645db43d56b402d49285a6ded4984bfcbcbc06582780dc05a80"
     }
    }
   },
@@ -585,6 +689,10 @@
     "ct": {
      "file": "data/bench/w055_ct.png",
      "sha256": "dab0bba53acdd7d658ef4a2bf147d8bcf81016faae4d96402e1b1593a2aea178"
+    },
+    "gpu": {
+     "file": "data/bench/w055_gpu.png",
+     "sha256": "dc8f4c5ba397db806b852cb0075a23c008df3bdd2c8b554eee477d2c9dbaf86a"
     }
    }
   },
@@ -602,6 +710,10 @@
     "ct": {
      "file": "data/bench/w056_ct.png",
      "sha256": "156992b325cb9b192ed76c07a80d904b34b4a49139b20e5e81b662b2387c3798"
+    },
+    "gpu": {
+     "file": "data/bench/w056_gpu.png",
+     "sha256": "c2a59ccce65176ac716e507dd584b59128174330f1df1f285013921c87142539"
     }
    }
   },
@@ -619,6 +731,10 @@
     "ct": {
      "file": "data/bench/w057_ct.png",
      "sha256": "14ef32a30434a86f24c3395c29d496d3602b336756304f5e32bc9c18148e1ac2"
+    },
+    "gpu": {
+     "file": "data/bench/w057_gpu.png",
+     "sha256": "0baae2f99ec2b4d643ddcb471b2605e9543a2e95e9dd37772c18f4abf513c263"
     }
    }
   },
@@ -636,6 +752,10 @@
     "ct": {
      "file": "data/bench/w058_ct.png",
      "sha256": "9a2d6eb4f994a44a02ccb7fe4b0134768b5b636da2faf21f6a1234ac24e79848"
+    },
+    "gpu": {
+     "file": "data/bench/w058_gpu.png",
+     "sha256": "c1547ad9c0c3909c8d8fc1a25150f3133235b84ee065cb55efe860719bc14164"
     }
    }
   },
@@ -653,6 +773,10 @@
     "ct": {
      "file": "data/bench/w059_ct.png",
      "sha256": "fead2cd274c88bd50672df8d0537bc3eca86c27387b0ceb21312864c4d1622b7"
+    },
+    "gpu": {
+     "file": "data/bench/w059_gpu.png",
+     "sha256": "f70d59d0ffde0448a8b1905418f972e0850b61416f105d7003d987bd64c6165d"
     }
    }
   }

docs/BENCH.md

@@ -16,9 +16,14 @@ Open it from the bar chart button in the [Unroll Space](https://huggingface.co/s
 
 > Every score is **agreement with another model's output**, the Vesuvius Challenge published ink detection on the 2.4 µm scan. It is not ground truth. A model can score lower because it is different, or even better. Use the numbers to find where detectors agree and disagree, then send the interesting regions to people who read Greek papyri.
 
+## Researcher takeaway
+On the public ink_9um run over all 37 windings (`docs/examples/bench/`), against the published 2.4 µm ink: median AP 0.319 for ink_9um, 0.117 for raw X-ray intensity and 0.089 for uniform noise, and ink_9um beats X-ray intensity on AP on every winding. On Pearson r the medians are 0.422, 0.266 and 0.000, but X-ray intensity outranks ink_9um on `w039`, `w053`, `w054` and `w057`. So: rank ink models by AP (threshold-free and robust to brightness), always report the X-ray floor next to your model, and look at windings where the floor wins before trusting a correlation. These are agreement scores with another model's output, not ground truth.
+
+The ink_9um model was trained by the Vesuvius Challenge and run by Chase Hendrick.
+
 ## Reference grid
 
-Each winding has a reference grid of 640 px on its long side (`data/bench/<id>_ink.png`, `<id>_ct.png`, and `<id>_gpu.png` for w046 to w051), lossless 8 bit PNG, with checksums and pixels per mm in `data/bench/index.json` (about 8.3 px per mm, so 1 px is about 120 µm on the papyrus). They are built by `tools/make_bench_refs.py` from the page's own layers. All metrics are computed on this grid, after your prediction is resampled onto it (bilinear, 2x2 mean pooled first while it is more than twice the grid).
+Each winding has a reference grid of 640 px on its long side (`data/bench/<id>_ink.png`, `<id>_ct.png`, and `<id>_gpu.png`, the public ink_9um run), lossless 8 bit PNG, with checksums and pixels per mm in `data/bench/index.json` (about 8.3 px per mm, so 1 px is about 120 µm on the papyrus). They are built by `tools/make_bench_refs.py` from the page's own layers. All metrics are computed on this grid, after your prediction is resampled onto it (bilinear, 2x2 mean pooled first while it is more than twice the grid).
 
 ## Python and CLI
 
@@ -88,7 +93,7 @@ B.core                                                                       //
 
 ## Leaderboard
 
-The Leaderboard tab loads any number of bundles from URLs (one per line, CORS) or files and ranks them per winding by Pearson r, AP, best F1, IoU or Dice, entirely in the browser. `?board=url1,url2` opens it preloaded. "Load examples" shows the public ink_9um runs next to two floors any ink model should beat: the X-ray intensity itself and uniform noise.
+The Leaderboard tab loads any number of bundles from URLs (one per line, CORS) or files and ranks them per winding by Pearson r, AP, best F1, IoU or Dice, entirely in the browser. `?board=url1,url2` opens it preloaded. "Load examples" shows the public ink_9um run on all 37 windings next to two floors any ink model should beat: the X-ray intensity itself and uniform noise.
 
 ## How it is tested
 

docs/examples/bench/w023-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:44Z",
+ "scroll": "PHerc0139",
+ "segment": "w023",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w023_ink.png",
+  "sha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d",
+  "width": 589,
+  "height": 640,
+  "pxPerMM": 9.861,
+  "alignTargetFile": "data/bench/w023_ink.png",
+  "alignTargetSha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d"
+ },
+ "prediction": {
+  "name": "w023_ct.png",
+  "sha256": "9e871edfb98d9c5455e3e8b6dae5bc2127b4b6c0ab2399ea5f80c9b9a38b1d2b",
+  "format": "png",
+  "width": 589,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.342535,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 376960,
+  "validFraction": 1.0,
+  "pearson": 0.342535,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 608.5,
+  "pearsonMatchedBlur": 0.464838,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.267945,
+  "dice": 0.422644,
+  "precision": 0.366091,
+  "recall": 0.499862,
+  "refPositiveFraction": 0.297854,
+  "predPositiveFraction": 0.40669,
+  "ap": 0.367097,
+  "bestF1": {
+   "f1": 0.528033,
+   "threshold": 0.04
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.413174,
+    "recall": 0.000615
+   },
+   {
+    "t": 0.98,
+    "precision": 0.419048,
+    "recall": 0.001568
+   },
+   {
+    "t": 0.96,
+    "precision": 0.393643,
+    "recall": 0.002868
+   },
+   {
+    "t": 0.94,
+    "precision": 0.371141,
+    "recall": 0.004605
+   },
+   {
+    "t": 0.92,
+    "precision": 0.381218,
+    "recall": 0.007303
+   },
+   {
+    "t": 0.9,
+    "precision": 0.379342,
+    "recall": 0.011186
+   },
+   {
+    "t": 0.88,
+    "precision": 0.369996,
+    "recall": 0.015969
+   },
+   {
+    "t": 0.86,
+    "precision": 0.372707,
+    "recall": 0.022622
+   },
+   {
+    "t": 0.84,
+    "precision": 0.374667,
+    "recall": 0.03135
+   },
+   {
+    "t": 0.82,
+    "precision": 0.376418,
+    "recall": 0.042849
+   },
+   {
+    "t": 0.8,
+    "precision": 0.374539,
+    "recall": 0.058827
+   },
+   {
+    "t": 0.78,
+    "precision": 0.374795,
+    "recall": 0.07541
+   },
+   {
+    "t": 0.76,
+    "precision": 0.376146,
+    "recall": 0.095022
+   },
+   {
+    "t": 0.74,
+    "precision": 0.37541,
+    "recall": 0.117128
+   },
+   {
+    "t": 0.72,
+    "precision": 0.37474,
+    "recall": 0.141246
+   },
+   {
+    "t": 0.7,
+    "precision": 0.37296,
+    "recall": 0.167342
+   },
+   {
+    "t": 0.68,
+    "precision": 0.372363,
+    "recall": 0.195745
+   },
+   {
+    "t": 0.66,
+    "precision": 0.371237,
+    "recall": 0.225608
+   },
+   {
+    "t": 0.64,
+    "precision": 0.37039,
+    "recall": 0.256531
+   },
+   {
+    "t": 0.62,
+    "precision": 0.369873,
+    "recall": 0.28952
+   },
+   {
+    "t": 0.6,
+    "precision": 0.369541,
+    "recall": 0.32951
+   },
+   {
+    "t": 0.58,
+    "precision": 0.368185,
+    "recall": 0.362721
+   },
+   {
+    "t": 0.56,
+    "precision": 0.367791,
+    "recall": 0.397118
+   },
+   {
+    "t": 0.54,
+    "precision": 0.36718,
+    "recall": 0.431336
+   },
+   {
+    "t": 0.52,
+    "precision": 0.366306,
+    "recall": 0.465439
+   },
+   {
+    "t": 0.5,
+    "precision": 0.366091,
+    "recall": 0.499862
+   },
+   {
+    "t": 0.48,
+    "precision": 0.366352,
+    "recall": 0.534802
+   },
+   {
+    "t": 0.46,
+    "precision": 0.366383,
+    "recall": 0.568842
+   },
+   {
+    "t": 0.44,
+    "precision": 0.366473,
+    "recall": 0.602437
+   },
+   {
+    "t": 0.42,
+    "precision": 0.365986,
+    "recall": 0.634233
+   },
+   {
+    "t": 0.4,
+    "precision": 0.365797,
+    "recall": 0.671889
+   },
+   {
+    "t": 0.38,
+    "precision": 0.366051,
+    "recall": 0.702411
+   },
+   {
+    "t": 0.36,
+    "precision": 0.365761,
+    "recall": 0.731054
+   },
+   {
+    "t": 0.34,
+    "precision": 0.365854,
+    "recall": 0.758984
+   },
+   {
+    "t": 0.32,
+    "precision": 0.365758,
+    "recall": 0.785
+   },
+   {
+    "t": 0.3,
+    "precision": 0.365498,
+    "recall": 0.808985
+   },
+   {
+    "t": 0.28,
+    "precision": 0.365094,
+    "recall": 0.830636
+   },
+   {
+    "t": 0.26,
+    "precision": 0.364846,
+    "recall": 0.851059
+   },
+   {
+    "t": 0.24,
+    "precision": 0.364451,
+    "recall": 0.869121
+   },
+   {
+    "t": 0.22,
+    "precision": 0.363978,
+    "recall": 0.885562
+   },
+   {
+    "t": 0.2,
+    "precision": 0.36361,
+    "recall": 0.903909
+   },
+   {
+    "t": 0.18,
+    "precision": 0.363437,
+    "recall": 0.917981
+   },
+   {
+    "t": 0.16,
+    "precision": 0.363229,
+    "recall": 0.930503
+   },
+   {
+    "t": 0.14,
+    "precision": 0.362969,
+    "recall": 0.941458
+   },
+   {
+    "t": 0.12,
+    "precision": 0.362743,
+    "recall": 0.951104
+   },
+   {
+    "t": 0.1,
+    "precision": 0.362424,
+    "recall": 0.959289
+   },
+   {
+    "t": 0.08,
+    "precision": 0.362018,
+    "recall": 0.966904
+   },
+   {
+    "t": 0.06,
+    "precision": 0.361708,
+    "recall": 0.973949
+   },
+   {
+    "t": 0.04,
+    "precision": 0.36137,
+    "recall": 0.980014
+   },
+   {
+    "t": 0.02,
+    "precision": 0.360408,
+    "recall": 0.985866
+   },
+   {
+    "t": 0.0,
+    "precision": 0.297854,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w023-ink9um.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:43Z",
+ "scroll": "PHerc0139",
+ "segment": "w023",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w023_ink.png",
+  "sha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d",
+  "width": 589,
+  "height": 640,
+  "pxPerMM": 9.861,
+  "alignTargetFile": "data/bench/w023_ink.png",
+  "alignTargetSha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d"
+ },
+ "prediction": {
+  "name": "w023_gpu.png",
+  "sha256": "14bedeb6a276f50d4783caec9774ccc4cf5899ddd67a0d57bf75dc6d04372d0c",
+  "format": "png",
+  "width": 589,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "scrollprize/ink_9um (seed 42, step 75000) on the 9.362 um scan",
+   "url": "https://huggingface.co/scrollprize/ink_9um"
+  }
+ },
+ "alignment": {
+  "method": "auto",
+  "target": "ink",
+  "tx": 0.563,
+  "ty": 0.5,
+  "sx": 1.0004,
+  "sy": 1.00157,
+  "ncc": 0.496932,
+  "coarseNcc": 0.676478,
+  "evals": 38395,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 376320,
+  "validFraction": 0.998302,
+  "pearson": 0.496932,
+  "blurSigmaPx": 2.5,
+  "blurSigmaUM": 253.5,
+  "pearsonMatchedBlur": 0.537678,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.322425,
+  "dice": 0.487627,
+  "precision": 0.611897,
+  "recall": 0.405312,
+  "refPositiveFraction": 0.29836,
+  "predPositiveFraction": 0.19763,
+  "ap": 0.572829,
+  "bestF1": {
+   "f1": 0.58117,
+   "threshold": 0.12
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.891379,
+    "recall": 0.004605
+   },
+   {
+    "t": 0.98,
+    "precision": 0.809298,
+    "recall": 0.069773
+   },
+   {
+    "t": 0.96,
+    "precision": 0.788433,
+    "recall": 0.091308
+   },
+   {
+    "t": 0.94,
+    "precision": 0.77342,
+    "recall": 0.109415
+   },
+   {
+    "t": 0.92,
+    "precision": 0.760949,
+    "recall": 0.12534
+   },
+   {
+    "t": 0.9,
+    "precision": 0.751885,
+    "recall": 0.140293
+   },
+   {
+    "t": 0.88,
+    "precision": 0.743538,
+    "recall": 0.154232
+   },
+   {
+    "t": 0.86,
+    "precision": 0.736346,
+    "recall": 0.168473
+   },
+   {
+    "t": 0.84,
+    "precision": 0.728813,
+    "recall": 0.181984
+   },
+   {
+    "t": 0.82,
+    "precision": 0.720644,
+    "recall": 0.195246
+   },
+   {
+    "t": 0.8,
+    "precision": 0.713955,
+    "recall": 0.209051
+   },
+   {
+    "t": 0.78,
+    "precision": 0.706978,
+    "recall": 0.222686
+   },
+   {
+    "t": 0.76,
+    "precision": 0.699595,
+    "recall": 0.235583
+   },
+   {
+    "t": 0.74,
+    "precision": 0.691579,
+    "recall": 0.248319
+   },
+   {
+    "t": 0.72,
+    "precision": 0.683979,
+    "recall": 0.260734
+   },
+   {
+    "t": 0.7,
+    "precision": 0.677081,
+    "recall": 0.273488
+   },
+   {
+    "t": 0.68,
+    "precision": 0.669728,
+    "recall": 0.28577
+   },
+   {
+    "t": 0.66,
+    "precision": 0.663324,
+    "recall": 0.298729
+   },
+   {
+    "t": 0.64,
+    "precision": 0.655696,
+    "recall": 0.311225
+   },
+   {
+    "t": 0.62,
+    "precision": 0.650146,
+    "recall": 0.325261
+   },
+   {
+    "t": 0.6,
+    "precision": 0.643842,
+    "recall": 0.338656
+   },
+   {
+    "t": 0.58,
+    "precision": 0.636991,
+    "recall": 0.351063
+   },
+   {
+    "t": 0.56,
+    "precision": 0.630731,
+    "recall": 0.364387
+   },
+   {
+    "t": 0.54,
+    "precision": 0.624137,
+    "recall": 0.378299
+   },
+   {
+    "t": 0.52,
+    "precision": 0.617467,
+    "recall": 0.392041
+   },
+   {
+    "t": 0.5,
+    "precision": 0.611897,
+    "recall": 0.405312
+   },
+   {
+    "t": 0.48,
+    "precision": 0.606205,
+    "recall": 0.419936
+   },
+   {
+    "t": 0.46,
+    "precision": 0.6,
+    "recall": 0.434747
+   },
+   {
+    "t": 0.44,
+    "precision": 0.594035,
+    "recall": 0.449149
+   },
+   {
+    "t": 0.42,
+    "precision": 0.587628,
+    "recall": 0.46404
+   },
+   {
+    "t": 0.4,
+    "precision": 0.581228,
+    "recall": 0.478896
+   },
+   {
+    "t": 0.38,
+    "precision": 0.57507,
+    "recall": 0.494269
+   },
+   {
+    "t": 0.36,
+    "precision": 0.56844,
+    "recall": 0.510122
+   },
+   {
+    "t": 0.34,
+    "precision": 0.561731,
+    "recall": 0.526269
+   },
+   {
+    "t": 0.32,
+    "precision": 0.554684,
+    "recall": 0.543405
+   },
+   {
+    "t": 0.3,
+    "precision": 0.548081,
+    "recall": 0.561841
+   },
+   {
+    "t": 0.28,
+    "precision": 0.540665,
+    "recall": 0.579645
+   },
+   {
+    "t": 0.26,
+    "precision": 0.53327,
+    "recall": 0.597868
+   },
+   {
+    "t": 0.24,
+    "precision": 0.525033,
+    "recall": 0.617177
+   },
+   {
+    "t": 0.22,
+    "precision": 0.517306,
+    "recall": 0.63816
+   },
+   {
+    "t": 0.2,
+    "precision": 0.509064,
+    "recall": 0.6595
+   },
+   {
+    "t": 0.18,
+    "precision": 0.500298,
+    "recall": 0.681232
+   },
+   {
+    "t": 0.16,
+    "precision": 0.49167,
+    "recall": 0.704442
+   },
+   {
+    "t": 0.14,
+    "precision": 0.482454,
+    "recall": 0.728907
+   },
+   {
+    "t": 0.12,
+    "precision": 0.472721,
+    "recall": 0.754193
+   },
+   {
+    "t": 0.1,
+    "precision": 0.461733,
+    "recall": 0.781384
+   },
+   {
+    "t": 0.08,
+    "precision": 0.449505,
+    "recall": 0.810169
+   },
+   {
+    "t": 0.06,
+    "precision": 0.435419,
+    "recall": 0.842161
+   },
+   {
+    "t": 0.04,
+    "precision": 0.418136,
+    "recall": 0.878882
+   },
+   {
+    "t": 0.02,
+    "precision": 0.394628,
+    "recall": 0.922425
+   },
+   {
+    "t": 0.0,
+    "precision": 0.29836,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w023-noise-baseline.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:45Z",
+ "scroll": "PHerc0139",
+ "segment": "w023",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w023_ink.png",
+  "sha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d",
+  "width": 589,
+  "height": 640,
+  "pxPerMM": 9.861,
+  "alignTargetFile": "data/bench/w023_ink.png",
+  "alignTargetSha256": "6e5a5d962eab758fe30776ac7db2ef8ea539c16c99938cc04bbaee5fa729839d"
+ },
+ "prediction": {
+  "name": "noise-seed23.png",
+  "sha256": "f66295cef27e849450ae58b39bb65cbfc7af90af40f3faad11e838e28f4e08f9",
+  "format": "png",
+  "width": 589,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: uniform noise",
+   "notes": "numpy default_rng(23).random, 8 bit"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.001026,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 376960,
+  "validFraction": 1.0,
+  "pearson": 0.001026,
+  "blurSigmaPx": 0.5,
+  "blurSigmaUM": 50.7,
+  "pearsonMatchedBlur": 0.001057,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.229363,
+  "dice": 0.373141,
+  "precision": 0.298139,
+  "recall": 0.498562,
+  "refPositiveFraction": 0.297854,
+  "predPositiveFraction": 0.498085,
+  "ap": 0.298344,
+  "bestF1": {
+   "f1": 0.458994,
+   "threshold": 0.0
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 1.0,
+    "recall": 0.0
+   },
+   {
+    "t": 0.98,
+    "precision": 0.296047,
+    "recall": 0.019612
+   },
+   {
+    "t": 0.96,
+    "precision": 0.296167,
+    "recall": 0.039437
+   },
+   {
+    "t": 0.94,
+    "precision": 0.298428,
+    "recall": 0.059521
+   },
+   {
+    "t": 0.92,
+    "precision": 0.300548,
+    "recall": 0.079579
+   },
+   {
+    "t": 0.9,
+    "precision": 0.300801,
+    "recall": 0.099645
+   },
+   {
+    "t": 0.88,
+    "precision": 0.298906,
+    "recall": 0.118749
+   },
+   {
+    "t": 0.86,
+    "precision": 0.299743,
+    "recall": 0.138289
+   },
+   {
+    "t": 0.84,
+    "precision": 0.299664,
+    "recall": 0.157875
+   },
+   {
+    "t": 0.82,
+    "precision": 0.299698,
+    "recall": 0.177602
+   },
+   {
+    "t": 0.8,
+    "precision": 0.300253,
+    "recall": 0.201881
+   },
+   {
+    "t": 0.78,
+    "precision": 0.299585,
+    "recall": 0.221003
+   },
+   {
+    "t": 0.76,
+    "precision": 0.298898,
+    "recall": 0.240116
+   },
+   {
+    "t": 0.74,
+    "precision": 0.299281,
+    "recall": 0.26053
+   },
+   {
+    "t": 0.72,
+    "precision": 0.299505,
+    "recall": 0.280275
+   },
+   {
+    "t": 0.7,
+    "precision": 0.299145,
+    "recall": 0.299566
+   },
+   {
+    "t": 0.68,
+    "precision": 0.29831,
+    "recall": 0.318813
+   },
+   {
+    "t": 0.66,
+    "precision": 0.297711,
+    "recall": 0.33773
+   },
+   {
+    "t": 0.64,
+    "precision": 0.297573,
+    "recall": 0.356825
+   },
+   {
+    "t": 0.62,
+    "precision": 0.29771,
+    "recall": 0.376526
+   },
+   {
+    "t": 0.6,
+    "precision": 0.298141,
+    "recall": 0.4002
+   },
+   {
+    "t": 0.58,
+    "precision": 0.298043,
+    "recall": 0.419517
+   },
+   {
+    "t": 0.56,
+    "precision": 0.297849,
+    "recall": 0.439201
+   },
+   {
+    "t": 0.54,
+    "precision": 0.297855,
+    "recall": 0.458973
+   },
+   {
+    "t": 0.52,
+    "precision": 0.297817,
+    "recall": 0.478398
+   },
+   {
+    "t": 0.5,
+    "precision": 0.298139,
+    "recall": 0.498562
+   },
+   {
+    "t": 0.48,
+    "precision": 0.297952,
+    "recall": 0.517612
+   },
+   {
+    "t": 0.46,
+    "precision": 0.29801,
+    "recall": 0.537171
+   },
+   {
+    "t": 0.44,
+    "precision": 0.298175,
+    "recall": 0.55664
+   },
+   {
+    "t": 0.42,
+    "precision": 0.29786,
+    "recall": 0.57545
+   },
+   {
+    "t": 0.4,
+    "precision": 0.298317,
+    "recall": 0.600059
+   },
+   {
+    "t": 0.38,
+    "precision": 0.298162,
+    "recall": 0.619359
+   },
+   {
+    "t": 0.36,
+    "precision": 0.298334,
+    "recall": 0.639487
+   },
+   {
+    "t": 0.34,
+    "precision": 0.298307,
+    "recall": 0.658814
+   },
+   {
+    "t": 0.32,
+    "precision": 0.298448,
+    "recall": 0.679103
+   },
+   {
+    "t": 0.3,
+    "precision": 0.29852,
+    "recall": 0.699009
+   },
+   {
+    "t": 0.28,
+    "precision": 0.298225,
+    "recall": 0.718211
+   },
+   {
+    "t": 0.26,
+    "precision": 0.298215,
+    "recall": 0.737698
+   },
+   {
+    "t": 0.24,
+    "precision": 0.298211,
+    "recall": 0.757354
+   },
+   {
+    "t": 0.22,
+    "precision": 0.298169,
+    "recall": 0.77669
+   },
+   {
+    "t": 0.2,
+    "precision": 0.298193,
+    "recall": 0.800123
+   },
+   {
+    "t": 0.18,
+    "precision": 0.298001,
+    "recall": 0.819209
+   },
+   {
+    "t": 0.16,
+    "precision": 0.297842,
+    "recall": 0.838536
+   },
+   {
+    "t": 0.14,
+    "precision": 0.297808,
+    "recall": 0.858308
+   },
+   {
+    "t": 0.12,
+    "precision": 0.297846,
+    "recall": 0.877893
+   },
+   {
+    "t": 0.1,
+    "precision": 0.297635,
+    "recall": 0.89722
+   },
+   {
+    "t": 0.08,
+    "precision": 0.297763,
+    "recall": 0.91734
+   },
+   {
+    "t": 0.06,
+    "precision": 0.297714,
+    "recall": 0.936551
+   },
+   {
+    "t": 0.04,
+    "precision": 0.297641,
+    "recall": 0.95635
+   },
+   {
+    "t": 0.02,
+    "precision": 0.29774,
+    "recall": 0.97622
+   },
+   {
+    "t": 0.0,
+    "precision": 0.297854,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w024-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:56Z",
+ "scroll": "PHerc0139",
+ "segment": "w024",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w024_ink.png",
+  "sha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327",
+  "width": 597,
+  "height": 640,
+  "pxPerMM": 9.6321,
+  "alignTargetFile": "data/bench/w024_ink.png",
+  "alignTargetSha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327"
+ },
+ "prediction": {
+  "name": "w024_ct.png",
+  "sha256": "7f21087fc1b5a1a5812e02d24134004524373408d5fe533939f12003f90c6517",
+  "format": "png",
+  "width": 597,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.361809,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 382080,
+  "validFraction": 1.0,
+  "pearson": 0.361809,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 622.9,
+  "pearsonMatchedBlur": 0.464513,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.288116,
+  "dice": 0.447345,
+  "precision": 0.418449,
+  "recall": 0.480526,
+  "refPositiveFraction": 0.333917,
+  "predPositiveFraction": 0.383454,
+  "ap": 0.41647,
+  "bestF1": {
+   "f1": 0.574501,
+   "threshold": 0.02
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.382353,
+    "recall": 0.000509
+   },
+   {
+    "t": 0.98,
+    "precision": 0.38412,
+    "recall": 0.001403
+   },
+   {
+    "t": 0.96,
+    "precision": 0.376357,
+    "recall": 0.002445
+   },
+   {
+    "t": 0.94,
+    "precision": 0.380223,
+    "recall": 0.00428
+   },
+   {
+    "t": 0.92,
+    "precision": 0.392257,
+    "recall": 0.006592
+   },
+   {
+    "t": 0.9,
+    "precision": 0.396662,
+    "recall": 0.009688
+   },
+   {
+    "t": 0.88,
+    "precision": 0.401505,
+    "recall": 0.013803
+   },
+   {
+    "t": 0.86,
+    "precision": 0.403503,
+    "recall": 0.01914
+   },
+   {
+    "t": 0.84,
+    "precision": 0.411504,
+    "recall": 0.026242
+   },
+   {
+    "t": 0.82,
+    "precision": 0.416875,
+    "recall": 0.03524
+   },
+   {
+    "t": 0.8,
+    "precision": 0.419535,
+    "recall": 0.048815
+   },
+   {
+    "t": 0.78,
+    "precision": 0.424343,
+    "recall": 0.063261
+   },
+   {
+    "t": 0.76,
+    "precision": 0.425433,
+    "recall": 0.080136
+   },
+   {
+    "t": 0.74,
+    "precision": 0.42628,
+    "recall": 0.099574
+   },
+   {
+    "t": 0.72,
+    "precision": 0.428473,
+    "recall": 0.122171
+   },
+   {
+    "t": 0.7,
+    "precision": 0.427396,
+    "recall": 0.14661
+   },
+   {
+    "t": 0.68,
+    "precision": 0.426788,
+    "recall": 0.173056
+   },
+   {
+    "t": 0.66,
+    "precision": 0.426028,
+    "recall": 0.202033
+   },
+   {
+    "t": 0.64,
+    "precision": 0.42663,
+    "recall": 0.233558
+   },
+   {
+    "t": 0.62,
+    "precision": 0.425587,
+    "recall": 0.265247
+   },
+   {
+    "t": 0.6,
+    "precision": 0.423207,
+    "recall": 0.305135
+   },
+   {
+    "t": 0.58,
+    "precision": 0.422423,
+    "recall": 0.339881
+   },
+   {
+    "t": 0.56,
+    "precision": 0.421306,
+    "recall": 0.374979
+   },
+   {
+    "t": 0.54,
+    "precision": 0.420412,
+    "recall": 0.409827
+   },
+   {
+    "t": 0.52,
+    "precision": 0.419469,
+    "recall": 0.445318
+   },
+   {
+    "t": 0.5,
+    "precision": 0.418449,
+    "recall": 0.480526
+   },
+   {
+    "t": 0.48,
+    "precision": 0.41824,
+    "recall": 0.516323
+   },
+   {
+    "t": 0.46,
+    "precision": 0.417336,
+    "recall": 0.550254
+   },
+   {
+    "t": 0.44,
+    "precision": 0.417034,
+    "recall": 0.584278
+   },
+   {
+    "t": 0.42,
+    "precision": 0.416464,
+    "recall": 0.617206
+   },
+   {
+    "t": 0.4,
+    "precision": 0.415636,
+    "recall": 0.655268
+   },
+   {
+    "t": 0.38,
+    "precision": 0.415486,
+    "recall": 0.68593
+   },
+   {
+    "t": 0.36,
+    "precision": 0.415521,
+    "recall": 0.715417
+   },
+   {
+    "t": 0.34,
+    "precision": 0.415132,
+    "recall": 0.742732
+   },
+   {
+    "t": 0.32,
+    "precision": 0.414668,
+    "recall": 0.768935
+   },
+   {
+    "t": 0.3,
+    "precision": 0.414072,
+    "recall": 0.793272
+   },
+   {
+    "t": 0.28,
+    "precision": 0.413718,
+    "recall": 0.816018
+   },
+   {
+    "t": 0.26,
+    "precision": 0.413501,
+    "recall": 0.837298
+   },
+   {
+    "t": 0.24,
+    "precision": 0.41271,
+    "recall": 0.856149
+   },
+   {
+    "t": 0.22,
+    "precision": 0.412062,
+    "recall": 0.873102
+   },
+   {
+    "t": 0.2,
+    "precision": 0.411404,
+    "recall": 0.891506
+   },
+   {
+    "t": 0.18,
+    "precision": 0.410778,
+    "recall": 0.904682
+   },
+   {
+    "t": 0.16,
+    "precision": 0.410245,
+    "recall": 0.917528
+   },
+   {
+    "t": 0.14,
+    "precision": 0.409727,
+    "recall": 0.928776
+   },
+   {
+    "t": 0.12,
+    "precision": 0.40926,
+    "recall": 0.939028
+   },
+   {
+    "t": 0.1,
+    "precision": 0.408687,
+    "recall": 0.948434
+   },
+   {
+    "t": 0.08,
+    "precision": 0.408117,
+    "recall": 0.95722
+   },
+   {
+    "t": 0.06,
+    "precision": 0.407705,
+    "recall": 0.965481
+   },
+   {
+    "t": 0.04,
+    "precision": 0.40722,
+    "recall": 0.97328
+   },
+   {
+    "t": 0.02,
+    "precision": 0.406165,
+    "recall": 0.981134
+   },
+   {
+    "t": 0.0,
+    "precision": 0.333917,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w024-ink9um.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:56Z",
+ "scroll": "PHerc0139",
+ "segment": "w024",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w024_ink.png",
+  "sha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327",
+  "width": 597,
+  "height": 640,
+  "pxPerMM": 9.6321,
+  "alignTargetFile": "data/bench/w024_ink.png",
+  "alignTargetSha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327"
+ },
+ "prediction": {
+  "name": "w024_gpu.png",
+  "sha256": "cc11fb74a30bc3fb48bd652946e3e9124d1a2d90947f613f715da37671a85efd",
+  "format": "png",
+  "width": 597,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "scrollprize/ink_9um (seed 42, step 75000) on the 9.362 um scan",
+   "url": "https://huggingface.co/scrollprize/ink_9um"
+  }
+ },
+ "alignment": {
+  "method": "auto",
+  "target": "ink",
+  "tx": 0.531,
+  "ty": 0.5,
+  "sx": 1.00018,
+  "sy": 1.00157,
+  "ncc": 0.572218,
+  "coarseNcc": 0.724818,
+  "evals": 40355,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 381440,
+  "validFraction": 0.998325,
+  "pearson": 0.572218,
+  "blurSigmaPx": 2.0,
+  "blurSigmaUM": 207.6,
+  "pearsonMatchedBlur": 0.608043,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.396013,
+  "dice": 0.567349,
+  "precision": 0.706089,
+  "recall": 0.474178,
+  "refPositiveFraction": 0.334477,
+  "predPositiveFraction": 0.22462,
+  "ap": 0.670837,
+  "bestF1": {
+   "f1": 0.644796,
+   "threshold": 0.14
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.932029,
+    "recall": 0.007093
+   },
+   {
+    "t": 0.98,
+    "precision": 0.874482,
+    "recall": 0.097638
+   },
+   {
+    "t": 0.96,
+    "precision": 0.86084,
+    "recall": 0.125675
+   },
+   {
+    "t": 0.94,
+    "precision": 0.851634,
+    "recall": 0.14748
+   },
+   {
+    "t": 0.92,
+    "precision": 0.843096,
+    "recall": 0.166401
+   },
+   {
+    "t": 0.9,
+    "precision": 0.83502,
+    "recall": 0.183677
+   },
+   {
+    "t": 0.88,
+    "precision": 0.827887,
+    "recall": 0.199972
+   },
+   {
+    "t": 0.86,
+    "precision": 0.820642,
+    "recall": 0.21582
+   },
+   {
+    "t": 0.84,
+    "precision": 0.813773,
+    "recall": 0.230814
+   },
+   {
+    "t": 0.82,
+    "precision": 0.807815,
+    "recall": 0.246796
+   },
+   {
+    "t": 0.8,
+    "precision": 0.801974,
+    "recall": 0.26172
+   },
+   {
+    "t": 0.78,
+    "precision": 0.795671,
+    "recall": 0.277169
+   },
+   {
+    "t": 0.76,
+    "precision": 0.790217,
+    "recall": 0.292382
+   },
+   {
+    "t": 0.74,
+    "precision": 0.783991,
+    "recall": 0.306836
+   },
+   {
+    "t": 0.72,
+    "precision": 0.777302,
+    "recall": 0.320333
+   },
+   {
+    "t": 0.7,
+    "precision": 0.771313,
+    "recall": 0.334496
+   },
+   {
+    "t": 0.68,
+    "precision": 0.764752,
+    "recall": 0.348824
+   },
+   {
+    "t": 0.66,
+    "precision": 0.757991,
+    "recall": 0.36225
+   },
+   {
+    "t": 0.64,
+    "precision": 0.751048,
+    "recall": 0.376328
+   },
+   {
+    "t": 0.62,
+    "precision": 0.744027,
+    "recall": 0.39006
+   },
+   {
+    "t": 0.6,
+    "precision": 0.737464,
+    "recall": 0.403925
+   },
+   {
+    "t": 0.58,
+    "precision": 0.730956,
+    "recall": 0.418167
+   },
+   {
+    "t": 0.56,
+    "precision": 0.724791,
+    "recall": 0.432228
+   },
+   {
+    "t": 0.54,
+    "precision": 0.719011,
+    "recall": 0.446415
+   },
+   {
+    "t": 0.52,
+    "precision": 0.71274,
+    "recall": 0.460516
+   },
+   {
+    "t": 0.5,
+    "precision": 0.706089,
+    "recall": 0.474178
+   },
+   {
+    "t": 0.48,
+    "precision": 0.699017,
+    "recall": 0.487761
+   },
+   {
+    "t": 0.46,
+    "precision": 0.69215,
+    "recall": 0.502065
+   },
+   {
+    "t": 0.44,
+    "precision": 0.684481,
+    "recall": 0.516605
+   },
+   {
+    "t": 0.42,
+    "precision": 0.676919,
+    "recall": 0.530619
+   },
+   {
+    "t": 0.4,
+    "precision": 0.669169,
+    "recall": 0.545026
+   },
+   {
+    "t": 0.38,
+    "precision": 0.66149,
+    "recall": 0.560216
+   },
+   {
+    "t": 0.36,
+    "precision": 0.654187,
+    "recall": 0.575751
+   },
+   {
+    "t": 0.34,
+    "precision": 0.646507,
+    "recall": 0.591192
+   },
+   {
+    "t": 0.32,
+    "precision": 0.638707,
+    "recall": 0.607032
+   },
+   {
+    "t": 0.3,
+    "precision": 0.630614,
+    "recall": 0.62321
+   },
+   {
+    "t": 0.28,
+    "precision": 0.622001,
+    "recall": 0.640015
+   },
+   {
+    "t": 0.26,
+    "precision": 0.613529,
+    "recall": 0.657313
+   },
+   {
+    "t": 0.24,
+    "precision": 0.604511,
+    "recall": 0.674471
+   },
+   {
+    "t": 0.22,
+    "precision": 0.594999,
+    "recall": 0.692655
+   },
+   {
+    "t": 0.2,
+    "precision": 0.585037,
+    "recall": 0.711153
+   },
+   {
+    "t": 0.18,
+    "precision": 0.57511,
+    "recall": 0.73005
+   },
+   {
+    "t": 0.16,
+    "precision": 0.564667,
+    "recall": 0.750288
+   },
+   {
+    "t": 0.14,
+    "precision": 0.553916,
+    "recall": 0.771349
+   },
+   {
+    "t": 0.12,
+    "precision": 0.542048,
+    "recall": 0.792927
+   },
+   {
+    "t": 0.1,
+    "precision": 0.528424,
+    "recall": 0.815281
+   },
+   {
+    "t": 0.08,
+    "precision": 0.513703,
+    "recall": 0.839924
+   },
+   {
+    "t": 0.06,
+    "precision": 0.496954,
+    "recall": 0.866887
+   },
+   {
+    "t": 0.04,
+    "precision": 0.476629,
+    "recall": 0.897479
+   },
+   {
+    "t": 0.02,
+    "precision": 0.448327,
+    "recall": 0.934451
+   },
+   {
+    "t": 0.0,
+    "precision": 0.334477,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w024-noise-baseline.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:47:58Z",
+ "scroll": "PHerc0139",
+ "segment": "w024",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w024_ink.png",
+  "sha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327",
+  "width": 597,
+  "height": 640,
+  "pxPerMM": 9.6321,
+  "alignTargetFile": "data/bench/w024_ink.png",
+  "alignTargetSha256": "496166c98593362e8fb6d3e10f489397028271d4f51714a05b7435aab685f327"
+ },
+ "prediction": {
+  "name": "noise-seed24.png",
+  "sha256": "0d3e3dbf7db544bf85db22e4904c2170188eff607aa99dba92f106800bd7cdc4",
+  "format": "png",
+  "width": 597,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: uniform noise",
+   "notes": "numpy default_rng(24).random, 8 bit"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": -0.000154,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 382080,
+  "validFraction": 1.0,
+  "pearson": -0.000154,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 622.9,
+  "pearsonMatchedBlur": 0.000533,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.249425,
+  "dice": 0.399263,
+  "precision": 0.333244,
+  "recall": 0.497903,
+  "refPositiveFraction": 0.333917,
+  "predPositiveFraction": 0.498909,
+  "ap": 0.334221,
+  "bestF1": {
+   "f1": 0.500656,
+   "threshold": 0.0
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 1.0,
+    "recall": 0.0
+   },
+   {
+    "t": 0.98,
+    "precision": 0.329879,
+    "recall": 0.019211
+   },
+   {
+    "t": 0.96,
+    "precision": 0.335803,
+    "recall": 0.039425
+   },
+   {
+    "t": 0.94,
+    "precision": 0.334846,
+    "recall": 0.058989
+   },
+   {
+    "t": 0.92,
+    "precision": 0.336679,
+    "recall": 0.079141
+   },
+   {
+    "t": 0.9,
+    "precision": 0.336394,
+    "recall": 0.098501
+   },
+   {
+    "t": 0.88,
+    "precision": 0.335897,
+    "recall": 0.118448
+   },
+   {
+    "t": 0.86,
+    "precision": 0.337719,
+    "recall": 0.139211
+   },
+   {
+    "t": 0.84,
+    "precision": 0.33856,
+    "recall": 0.159598
+   },
+   {
+    "t": 0.82,
+    "precision": 0.337444,
+    "recall": 0.179107
+   },
+   {
+    "t": 0.8,
+    "precision": 0.335787,
+    "recall": 0.202339
+   },
+   {
+    "t": 0.78,
+    "precision": 0.335545,
+    "recall": 0.22155
+   },
+   {
+    "t": 0.76,
+    "precision": 0.335695,
+    "recall": 0.241411
+   },
+   {
+    "t": 0.74,
+    "precision": 0.335913,
+    "recall": 0.261273
+   },
+   {
+    "t": 0.72,
+    "precision": 0.335316,
+    "recall": 0.280531
+   },
+   {
+    "t": 0.7,
+    "precision": 0.33509,
+    "recall": 0.299954
+   },
+   {
+    "t": 0.68,
+    "precision": 0.334654,
+    "recall": 0.31922
+   },
+   {
+    "t": 0.66,
+    "precision": 0.334034,
+    "recall": 0.338713
+   },
+   {
+    "t": 0.64,
+    "precision": 0.334197,
+    "recall": 0.3579
+   },
+   {
+    "t": 0.62,
+    "precision": 0.333655,
+    "recall": 0.377229
+   },
+   {
+    "t": 0.6,
+    "precision": 0.333989,
+    "recall": 0.400618
+   },
+   {
+    "t": 0.58,
+    "precision": 0.333871,
+    "recall": 0.420079
+   },
+   {
+    "t": 0.56,
+    "precision": 0.3337,
+    "recall": 0.439651
+   },
+   {
+    "t": 0.54,
+    "precision": 0.333426,
+    "recall": 0.459089
+   },
+   {
+    "t": 0.52,
+    "precision": 0.33325,
+    "recall": 0.478434
+   },
+   {
+    "t": 0.5,
+    "precision": 0.333244,
+    "recall": 0.497903
+   },
+   {
+    "t": 0.48,
+    "precision": 0.332977,
+    "recall": 0.517052
+   },
+   {
+    "t": 0.46,
+    "precision": 0.333027,
+    "recall": 0.5366
+   },
+   {
+    "t": 0.44,
+    "precision": 0.333189,
+    "recall": 0.556649
+   },
+   {
+    "t": 0.42,
+    "precision": 0.333277,
+    "recall": 0.576519
+   },
+   {
+    "t": 0.4,
+    "precision": 0.333066,
+    "recall": 0.599375
+   },
+   {
+    "t": 0.38,
+    "precision": 0.333098,
+    "recall": 0.618695
+   },
+   {
+    "t": 0.36,
+    "precision": 0.333016,
+    "recall": 0.638126
+   },
+   {
+    "t": 0.34,
+    "precision": 0.333275,
+    "recall": 0.658701
+   },
+   {
+    "t": 0.32,
+    "precision": 0.333309,
+    "recall": 0.678264
+   },
+   {
+    "t": 0.3,
+    "precision": 0.333411,
+    "recall": 0.697867
+   },
+   {
+    "t": 0.28,
+    "precision": 0.333264,
+    "recall": 0.71689
+   },
+   {
+    "t": 0.26,
+    "precision": 0.333387,
+    "recall": 0.736462
+   },
+   {
+    "t": 0.24,
+    "precision": 0.333415,
+    "recall": 0.756378
+   },
+   {
+    "t": 0.22,
+    "precision": 0.33365,
+    "recall": 0.776412
+   },
+   {
+    "t": 0.2,
+    "precision": 0.333478,
+    "recall": 0.799393
+   },
+   {
+    "t": 0.18,
+    "precision": 0.33341,
+    "recall": 0.818667
+   },
+   {
+    "t": 0.16,
+    "precision": 0.333651,
+    "recall": 0.83856
+   },
+   {
+    "t": 0.14,
+    "precision": 0.333609,
+    "recall": 0.858179
+   },
+   {
+    "t": 0.12,
+    "precision": 0.333629,
+    "recall": 0.877554
+   },
+   {
+    "t": 0.1,
+    "precision": 0.333554,
+    "recall": 0.897188
+   },
+   {
+    "t": 0.08,
+    "precision": 0.333709,
+    "recall": 0.917552
+   },
+   {
+    "t": 0.06,
+    "precision": 0.333833,
+    "recall": 0.937225
+   },
+   {
+    "t": 0.04,
+    "precision": 0.333941,
+    "recall": 0.957063
+   },
+   {
+    "t": 0.02,
+    "precision": 0.333927,
+    "recall": 0.976752
+   },
+   {
+    "t": 0.0,
+    "precision": 0.333917,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w025-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:10Z",
+ "scroll": "PHerc0139",
+ "segment": "w025",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w025_ink.png",
+  "sha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8",
+  "width": 598,
+  "height": 640,
+  "pxPerMM": 9.3937,
+  "alignTargetFile": "data/bench/w025_ink.png",
+  "alignTargetSha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8"
+ },
+ "prediction": {
+  "name": "w025_ct.png",
+  "sha256": "e30f050d29ccc83eaf9b29beafe66f0a47a35ac2a7f30dbd525f462c9ea325fa",
+  "format": "png",
+  "width": 598,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.359953,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 382720,
+  "validFraction": 1.0,
+  "pearson": 0.359953,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 638.7,
+  "pearsonMatchedBlur": 0.437292,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.204451,
+  "dice": 0.339493,
+  "precision": 0.236587,
+  "recall": 0.600833,
+  "refPositiveFraction": 0.155516,
+  "predPositiveFraction": 0.394947,
+  "ap": 0.242417,
+  "bestF1": {
+   "f1": 0.343571,
+   "threshold": 0.44
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.259459,
+    "recall": 0.000806
+   },
+   {
+    "t": 0.98,
+    "precision": 0.320594,
+    "recall": 0.002537
+   },
+   {
+    "t": 0.96,
+    "precision": 0.293318,
+    "recall": 0.004352
+   },
+   {
+    "t": 0.94,
+    "precision": 0.300469,
+    "recall": 0.007527
+   },
+   {
+    "t": 0.92,
+    "precision": 0.283003,
+    "recall": 0.011526
+   },
+   {
+    "t": 0.9,
+    "precision": 0.287168,
+    "recall": 0.018011
+   },
+   {
+    "t": 0.88,
+    "precision": 0.287614,
+    "recall": 0.026529
+   },
+   {
+    "t": 0.86,
+    "precision": 0.286919,
+    "recall": 0.037148
+   },
+   {
+    "t": 0.84,
+    "precision": 0.287639,
+    "recall": 0.050942
+   },
+   {
+    "t": 0.82,
+    "precision": 0.287651,
+    "recall": 0.068449
+   },
+   {
+    "t": 0.8,
+    "precision": 0.282144,
+    "recall": 0.091433
+   },
+   {
+    "t": 0.78,
+    "precision": 0.27808,
+    "recall": 0.114266
+   },
+   {
+    "t": 0.76,
+    "precision": 0.275642,
+    "recall": 0.141283
+   },
+   {
+    "t": 0.74,
+    "precision": 0.273569,
+    "recall": 0.170349
+   },
+   {
+    "t": 0.72,
+    "precision": 0.271296,
+    "recall": 0.20296
+   },
+   {
+    "t": 0.7,
+    "precision": 0.268313,
+    "recall": 0.236563
+   },
+   {
+    "t": 0.68,
+    "precision": 0.265416,
+    "recall": 0.272266
+   },
+   {
+    "t": 0.66,
+    "precision": 0.26319,
+    "recall": 0.309935
+   },
+   {
+    "t": 0.64,
+    "precision": 0.259919,
+    "recall": 0.34725
+   },
+   {
+    "t": 0.62,
+    "precision": 0.257365,
+    "recall": 0.38544
+   },
+   {
+    "t": 0.6,
+    "precision": 0.253624,
+    "recall": 0.430652
+   },
+   {
+    "t": 0.58,
+    "precision": 0.249869,
+    "recall": 0.466422
+   },
+   {
+    "t": 0.56,
+    "precision": 0.24668,
+    "recall": 0.502159
+   },
+   {
+    "t": 0.54,
+    "precision": 0.243306,
+    "recall": 0.536501
+   },
+   {
+    "t": 0.52,
+    "precision": 0.239897,
+    "recall": 0.569129
+   },
+   {
+    "t": 0.5,
+    "precision": 0.236587,
+    "recall": 0.600833
+   },
+   {
+    "t": 0.48,
+    "precision": 0.233961,
+    "recall": 0.632302
+   },
+   {
+    "t": 0.46,
+    "precision": 0.231197,
+    "recall": 0.662108
+   },
+   {
+    "t": 0.44,
+    "precision": 0.228708,
+    "recall": 0.690217
+   },
+   {
+    "t": 0.42,
+    "precision": 0.225623,
+    "recall": 0.715251
+   },
+   {
+    "t": 0.4,
+    "precision": 0.222726,
+    "recall": 0.74472
+   },
+   {
+    "t": 0.38,
+    "precision": 0.220147,
+    "recall": 0.767167
+   },
+   {
+    "t": 0.36,
+    "precision": 0.217627,
+    "recall": 0.787631
+   },
+   {
+    "t": 0.34,
+    "precision": 0.215409,
+    "recall": 0.807238
+   },
+   {
+    "t": 0.32,
+    "precision": 0.213241,
+    "recall": 0.825467
+   },
+   {
+    "t": 0.3,
+    "precision": 0.211177,
+    "recall": 0.842319
+   },
+   {
+    "t": 0.28,
+    "precision": 0.209103,
+    "recall": 0.857172
+   },
+   {
+    "t": 0.26,
+    "precision": 0.20715,
+    "recall": 0.870781
+   },
+   {
+    "t": 0.24,
+    "precision": 0.20542,
+    "recall": 0.883936
+   },
+   {
+    "t": 0.22,
+    "precision": 0.203667,
+    "recall": 0.895193
+   },
+   {
+    "t": 0.2,
+    "precision": 0.201809,
+    "recall": 0.908701
+   },
+   {
+    "t": 0.18,
+    "precision": 0.200306,
+    "recall": 0.919017
+   },
+   {
+    "t": 0.16,
+    "precision": 0.198907,
+    "recall": 0.928443
+   },
+   {
+    "t": 0.14,
+    "precision": 0.197644,
+    "recall": 0.937953
+   },
+   {
+    "t": 0.12,
+    "precision": 0.196428,
+    "recall": 0.946588
+   },
+   {
+    "t": 0.1,
+    "precision": 0.195035,
+    "recall": 0.954065
+   },
+   {
+    "t": 0.08,
+    "precision": 0.193604,
+    "recall": 0.961038
+   },
+   {
+    "t": 0.06,
+    "precision": 0.192206,
+    "recall": 0.968111
+   },
+   {
+    "t": 0.04,
+    "precision": 0.191009,
+    "recall": 0.975554
+   },
+   {
+    "t": 0.02,
+    "precision": 0.189517,
+    "recall": 0.982661
+   },
+   {
+    "t": 0.0,
+    "precision": 0.155516,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w025-ink9um.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:09Z",
+ "scroll": "PHerc0139",
+ "segment": "w025",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w025_ink.png",
+  "sha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8",
+  "width": 598,
+  "height": 640,
+  "pxPerMM": 9.3937,
+  "alignTargetFile": "data/bench/w025_ink.png",
+  "alignTargetSha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8"
+ },
+ "prediction": {
+  "name": "w025_gpu.png",
+  "sha256": "c129d596d1fe215b2a5c671eff5143e938bd4c4d34ac1a6c7b9843d1f263f226",
+  "format": "png",
+  "width": 598,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "scrollprize/ink_9um (seed 42, step 75000) on the 9.362 um scan",
+   "url": "https://huggingface.co/scrollprize/ink_9um"
+  }
+ },
+ "alignment": {
+  "method": "auto",
+  "target": "ink",
+  "tx": 0.531,
+  "ty": 0.594,
+  "sx": 1.00403,
+  "sy": 1.00007,
+  "ncc": 0.592898,
+  "coarseNcc": 0.749308,
+  "evals": 40350,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 382122,
+  "validFraction": 0.998438,
+  "pearson": 0.592898,
+  "blurSigmaPx": 2.0,
+  "blurSigmaUM": 212.9,
+  "pearsonMatchedBlur": 0.623003,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.36052,
+  "dice": 0.529974,
+  "precision": 0.61375,
+  "recall": 0.466322,
+  "refPositiveFraction": 0.155759,
+  "predPositiveFraction": 0.118344,
+  "ap": 0.562828,
+  "bestF1": {
+   "f1": 0.547703,
+   "threshold": 0.34
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.968254,
+    "recall": 0.006149
+   },
+   {
+    "t": 0.98,
+    "precision": 0.85798,
+    "recall": 0.110637
+   },
+   {
+    "t": 0.96,
+    "precision": 0.843237,
+    "recall": 0.13972
+   },
+   {
+    "t": 0.94,
+    "precision": 0.829066,
+    "recall": 0.161595
+   },
+   {
+    "t": 0.92,
+    "precision": 0.816653,
+    "recall": 0.180279
+   },
+   {
+    "t": 0.9,
+    "precision": 0.806396,
+    "recall": 0.197836
+   },
+   {
+    "t": 0.88,
+    "precision": 0.794865,
+    "recall": 0.214318
+   },
+   {
+    "t": 0.86,
+    "precision": 0.783196,
+    "recall": 0.22991
+   },
+   {
+    "t": 0.84,
+    "precision": 0.773132,
+    "recall": 0.243855
+   },
+   {
+    "t": 0.82,
+    "precision": 0.763133,
+    "recall": 0.258472
+   },
+   {
+    "t": 0.8,
+    "precision": 0.755407,
+    "recall": 0.272888
+   },
+   {
+    "t": 0.78,
+    "precision": 0.746371,
+    "recall": 0.285976
+   },
+   {
+    "t": 0.76,
+    "precision": 0.737928,
+    "recall": 0.300408
+   },
+   {
+    "t": 0.74,
+    "precision": 0.728903,
+    "recall": 0.314185
+   },
+   {
+    "t": 0.72,
+    "precision": 0.720131,
+    "recall": 0.327912
+   },
+   {
+    "t": 0.7,
+    "precision": 0.71035,
+    "recall": 0.341336
+   },
+   {
+    "t": 0.68,
+    "precision": 0.701014,
+    "recall": 0.354223
+   },
+   {
+    "t": 0.66,
+    "precision": 0.691092,
+    "recall": 0.367311
+   },
+   {
+    "t": 0.64,
+    "precision": 0.682133,
+    "recall": 0.380131
+   },
+   {
+    "t": 0.62,
+    "precision": 0.671493,
+    "recall": 0.392715
+   },
+   {
+    "t": 0.6,
+    "precision": 0.661459,
+    "recall": 0.405484
+   },
+   {
+    "t": 0.58,
+    "precision": 0.652231,
+    "recall": 0.417262
+   },
+   {
+    "t": 0.56,
+    "precision": 0.64251,
+    "recall": 0.429157
+   },
+   {
+    "t": 0.54,
+    "precision": 0.632915,
+    "recall": 0.440985
+   },
+   {
+    "t": 0.52,
+    "precision": 0.623577,
+    "recall": 0.45362
+   },
+   {
+    "t": 0.5,
+    "precision": 0.61375,
+    "recall": 0.466322
+   },
+   {
+    "t": 0.48,
+    "precision": 0.604075,
+    "recall": 0.479696
+   },
+   {
+    "t": 0.46,
+    "precision": 0.593219,
+    "recall": 0.492683
+   },
+   {
+    "t": 0.44,
+    "precision": 0.581058,
+    "recall": 0.504763
+   },
+   {
+    "t": 0.42,
+    "precision": 0.570364,
+    "recall": 0.518271
+   },
+   {
+    "t": 0.4,
+    "precision": 0.560299,
+    "recall": 0.531427
+   },
+   {
+    "t": 0.38,
+    "precision": 0.548314,
+    "recall": 0.544095
+   },
+   {
+    "t": 0.36,
+    "precision": 0.536368,
+    "recall": 0.557654
+   },
+   {
+    "t": 0.34,
+    "precision": 0.525257,
+    "recall": 0.572153
+   },
+   {
+    "t": 0.32,
+    "precision": 0.513583,
+    "recall": 0.586351
+   },
+   {
+    "t": 0.3,
+    "precision": 0.50199,
+    "recall": 0.601892
+   },
+   {
+    "t": 0.28,
+    "precision": 0.490008,
+    "recall": 0.617131
+   },
+   {
+    "t": 0.26,
+    "precision": 0.477762,
+    "recall": 0.632773
+   },
+   {
+    "t": 0.24,
+    "precision": 0.465209,
+    "recall": 0.649944
+   },
+   {
+    "t": 0.22,
+    "precision": 0.451807,
+    "recall": 0.666426
+   },
+   {
+    "t": 0.2,
+    "precision": 0.438308,
+    "recall": 0.684823
+   },
+   {
+    "t": 0.18,
+    "precision": 0.424135,
+    "recall": 0.702549
+   },
+   {
+    "t": 0.16,
+    "precision": 0.40955,
+    "recall": 0.722072
+   },
+   {
+    "t": 0.14,
+    "precision": 0.393667,
+    "recall": 0.743191
+   },
+   {
+    "t": 0.12,
+    "precision": 0.376616,
+    "recall": 0.764596
+   },
+   {
+    "t": 0.1,
+    "precision": 0.358264,
+    "recall": 0.7879
+   },
+   {
+    "t": 0.08,
+    "precision": 0.338383,
+    "recall": 0.815168
+   },
+   {
+    "t": 0.06,
+    "precision": 0.315728,
+    "recall": 0.844184
+   },
+   {
+    "t": 0.04,
+    "precision": 0.288123,
+    "recall": 0.879181
+   },
+   {
+    "t": 0.02,
+    "precision": 0.252037,
+    "recall": 0.922294
+   },
+   {
+    "t": 0.0,
+    "precision": 0.155759,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w025-noise-baseline.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:11Z",
+ "scroll": "PHerc0139",
+ "segment": "w025",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w025_ink.png",
+  "sha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8",
+  "width": 598,
+  "height": 640,
+  "pxPerMM": 9.3937,
+  "alignTargetFile": "data/bench/w025_ink.png",
+  "alignTargetSha256": "298aee5988b1120978977f573824bbb1d550fd877a8b233dabea7724378758a8"
+ },
+ "prediction": {
+  "name": "noise-seed25.png",
+  "sha256": "cca11cae5b62ce76bbf1dc053614ab189878680be8021c962628d229ffceb639",
+  "format": "png",
+  "width": 598,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: uniform noise",
+   "notes": "numpy default_rng(25).random, 8 bit"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.000355,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 382720,
+  "validFraction": 1.0,
+  "pearson": 0.000355,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 638.7,
+  "pearsonMatchedBlur": 0.000806,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.134334,
+  "dice": 0.236851,
+  "precision": 0.155441,
+  "recall": 0.497303,
+  "refPositiveFraction": 0.155516,
+  "predPositiveFraction": 0.497541,
+  "ap": 0.156205,
+  "bestF1": {
+   "f1": 0.269171,
+   "threshold": 0.0
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 1.0,
+    "recall": 0.0
+   },
+   {
+    "t": 0.98,
+    "precision": 0.167365,
+    "recall": 0.021489
+   },
+   {
+    "t": 0.96,
+    "precision": 0.161275,
+    "recall": 0.040794
+   },
+   {
+    "t": 0.94,
+    "precision": 0.159883,
+    "recall": 0.060451
+   },
+   {
+    "t": 0.92,
+    "precision": 0.159164,
+    "recall": 0.080378
+   },
+   {
+    "t": 0.9,
+    "precision": 0.159305,
+    "recall": 0.10022
+   },
+   {
+    "t": 0.88,
+    "precision": 0.158096,
+    "recall": 0.119323
+   },
+   {
+    "t": 0.86,
+    "precision": 0.157868,
+    "recall": 0.138964
+   },
+   {
+    "t": 0.84,
+    "precision": 0.157373,
+    "recall": 0.15847
+   },
+   {
+    "t": 0.82,
+    "precision": 0.156813,
+    "recall": 0.177893
+   },
+   {
+    "t": 0.8,
+    "precision": 0.155958,
+    "recall": 0.200272
+   },
+   {
+    "t": 0.78,
+    "precision": 0.155983,
+    "recall": 0.219829
+   },
+   {
+    "t": 0.76,
+    "precision": 0.155989,
+    "recall": 0.239436
+   },
+   {
+    "t": 0.74,
+    "precision": 0.155477,
+    "recall": 0.258186
+   },
+   {
+    "t": 0.72,
+    "precision": 0.155227,
+    "recall": 0.277407
+   },
+   {
+    "t": 0.7,
+    "precision": 0.15537,
+    "recall": 0.297233
+   },
+   {
+    "t": 0.68,
+    "precision": 0.155284,
+    "recall": 0.316605
+   },
+   {
+    "t": 0.66,
+    "precision": 0.154942,
+    "recall": 0.335607
+   },
+   {
+    "t": 0.64,
+    "precision": 0.155212,
+    "recall": 0.355769
+   },
+   {
+    "t": 0.62,
+    "precision": 0.155133,
+    "recall": 0.374603
+   },
+   {
+    "t": 0.6,
+    "precision": 0.155409,
+    "recall": 0.399099
+   },
+   {
+    "t": 0.58,
+    "precision": 0.155281,
+    "recall": 0.418505
+   },
+   {
+    "t": 0.56,
+    "precision": 0.155247,
+    "recall": 0.438028
+   },
+   {
+    "t": 0.54,
+    "precision": 0.155428,
+    "recall": 0.458055
+   },
+   {
+    "t": 0.52,
+    "precision": 0.155539,
+    "recall": 0.478167
+   },
+   {
+    "t": 0.5,
+    "precision": 0.155441,
+    "recall": 0.497303
+   },
+   {
+    "t": 0.48,
+    "precision": 0.15554,
+    "recall": 0.517146
+   },
+   {
+    "t": 0.46,
+    "precision": 0.155535,
+    "recall": 0.536887
+   },
+   {
+    "t": 0.44,
+    "precision": 0.155541,
+    "recall": 0.556579
+   },
+   {
+    "t": 0.42,
+    "precision": 0.155464,
+    "recall": 0.575883
+   },
+   {
+    "t": 0.4,
+    "precision": 0.155432,
+    "recall": 0.59912
+   },
+   {
+    "t": 0.38,
+    "precision": 0.155575,
+    "recall": 0.6195
+   },
+   {
+    "t": 0.36,
+    "precision": 0.155318,
+    "recall": 0.638065
+   },
+   {
+    "t": 0.34,
+    "precision": 0.155339,
+    "recall": 0.657286
+   },
+   {
+    "t": 0.32,
+    "precision": 0.155334,
+    "recall": 0.676742
+   },
+   {
+    "t": 0.3,
+    "precision": 0.155558,
+    "recall": 0.697475
+   },
+   {
+    "t": 0.28,
+    "precision": 0.155358,
+    "recall": 0.716275
+   },
+   {
+    "t": 0.26,
+    "precision": 0.155151,
+    "recall": 0.734908
+   },
+   {
+    "t": 0.24,
+    "precision": 0.155166,
+    "recall": 0.754633
+   },
+   {
+    "t": 0.22,
+    "precision": 0.155302,
+    "recall": 0.77466
+   },
+   {
+    "t": 0.2,
+    "precision": 0.155375,
+    "recall": 0.798501
+   },
+   {
+    "t": 0.18,
+    "precision": 0.155477,
+    "recall": 0.818982
+   },
+   {
+    "t": 0.16,
+    "precision": 0.155276,
+    "recall": 0.837296
+   },
+   {
+    "t": 0.14,
+    "precision": 0.155432,
+    "recall": 0.857693
+   },
+   {
+    "t": 0.12,
+    "precision": 0.15544,
+    "recall": 0.877468
+   },
+   {
+    "t": 0.1,
+    "precision": 0.15553,
+    "recall": 0.897327
+   },
+   {
+    "t": 0.08,
+    "precision": 0.155506,
+    "recall": 0.917321
+   },
+   {
+    "t": 0.06,
+    "precision": 0.155471,
+    "recall": 0.936961
+   },
+   {
+    "t": 0.04,
+    "precision": 0.155464,
+    "recall": 0.956232
+   },
+   {
+    "t": 0.02,
+    "precision": 0.155598,
+    "recall": 0.976512
+   },
+   {
+    "t": 0.0,
+    "precision": 0.155516,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w026-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:29Z",
+ "scroll": "PHerc0139",
+ "segment": "w026",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w026_ink.png",
+  "sha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791",
+  "width": 629,
+  "height": 640,
+  "pxPerMM": 9.1785,
+  "alignTargetFile": "data/bench/w026_ink.png",
+  "alignTargetSha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791"
+ },
+ "prediction": {
+  "name": "w026_ct.png",
+  "sha256": "d005b08b88e31ad0cc24bfbbff91041e8abad6275d39cbb243dddfe78b46f3f0",
+  "format": "png",
+  "width": 629,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.278906,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 402560,
+  "validFraction": 1.0,
+  "pearson": 0.278906,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 653.7,
+  "pearsonMatchedBlur": 0.376576,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.064599,
+  "dice": 0.121359,
+  "precision": 0.068329,
+  "recall": 0.542037,
+  "refPositiveFraction": 0.048634,
+  "predPositiveFraction": 0.385801,
+  "ap": 0.072337,
+  "bestF1": {
+   "f1": 0.122104,
+   "threshold": 0.6
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.225806,
+    "recall": 0.002503
+   },
+   {
+    "t": 0.98,
+    "precision": 0.170956,
+    "recall": 0.00475
+   },
+   {
+    "t": 0.96,
+    "precision": 0.152941,
+    "recall": 0.007968
+   },
+   {
+    "t": 0.94,
+    "precision": 0.142022,
+    "recall": 0.012412
+   },
+   {
+    "t": 0.92,
+    "precision": 0.133232,
+    "recall": 0.017979
+   },
+   {
+    "t": 0.9,
+    "precision": 0.125635,
+    "recall": 0.025283
+   },
+   {
+    "t": 0.88,
+    "precision": 0.112322,
+    "recall": 0.032639
+   },
+   {
+    "t": 0.86,
+    "precision": 0.106697,
+    "recall": 0.04362
+   },
+   {
+    "t": 0.84,
+    "precision": 0.101162,
+    "recall": 0.056901
+   },
+   {
+    "t": 0.82,
+    "precision": 0.095673,
+    "recall": 0.072173
+   },
+   {
+    "t": 0.8,
+    "precision": 0.090627,
+    "recall": 0.094034
+   },
+   {
+    "t": 0.78,
+    "precision": 0.086661,
+    "recall": 0.113393
+   },
+   {
+    "t": 0.76,
+    "precision": 0.084308,
+    "recall": 0.13699
+   },
+   {
+    "t": 0.74,
+    "precision": 0.082699,
+    "recall": 0.163142
+   },
+   {
+    "t": 0.72,
+    "precision": 0.080023,
+    "recall": 0.188477
+   },
+   {
+    "t": 0.7,
+    "precision": 0.079188,
+    "recall": 0.218868
+   },
+   {
+    "t": 0.68,
+    "precision": 0.07806,
+    "recall": 0.251149
+   },
+   {
+    "t": 0.66,
+    "precision": 0.076513,
+    "recall": 0.281285
+   },
+   {
+    "t": 0.64,
+    "precision": 0.075215,
+    "recall": 0.313004
+   },
+   {
+    "t": 0.62,
+    "precision": 0.073916,
+    "recall": 0.343855
+   },
+   {
+    "t": 0.6,
+    "precision": 0.07264,
+    "recall": 0.382726
+   },
+   {
+    "t": 0.58,
+    "precision": 0.071338,
+    "recall": 0.413525
+   },
+   {
+    "t": 0.56,
+    "precision": 0.070653,
+    "recall": 0.44693
+   },
+   {
+    "t": 0.54,
+    "precision": 0.069849,
+    "recall": 0.479416
+   },
+   {
+    "t": 0.52,
+    "precision": 0.069073,
+    "recall": 0.510982
+   },
+   {
+    "t": 0.5,
+    "precision": 0.068329,
+    "recall": 0.542037
+   },
+   {
+    "t": 0.48,
+    "precision": 0.067272,
+    "recall": 0.569517
+   },
+   {
+    "t": 0.46,
+    "precision": 0.066605,
+    "recall": 0.598325
+   },
+   {
+    "t": 0.44,
+    "precision": 0.065942,
+    "recall": 0.625396
+   },
+   {
+    "t": 0.42,
+    "precision": 0.065164,
+    "recall": 0.650577
+   },
+   {
+    "t": 0.4,
+    "precision": 0.064333,
+    "recall": 0.679436
+   },
+   {
+    "t": 0.38,
+    "precision": 0.063741,
+    "recall": 0.703136
+   },
+   {
+    "t": 0.36,
+    "precision": 0.063055,
+    "recall": 0.723669
+   },
+   {
+    "t": 0.34,
+    "precision": 0.062479,
+    "recall": 0.744305
+   },
+   {
+    "t": 0.32,
+    "precision": 0.062059,
+    "recall": 0.765451
+   },
+   {
+    "t": 0.3,
+    "precision": 0.061548,
+    "recall": 0.783686
+   },
+   {
+    "t": 0.28,
+    "precision": 0.06097,
+    "recall": 0.800184
+   },
+   {
+    "t": 0.26,
+    "precision": 0.060372,
+    "recall": 0.814588
+   },
+   {
+    "t": 0.24,
+    "precision": 0.059949,
+    "recall": 0.830064
+   },
+   {
+    "t": 0.22,
+    "precision": 0.059607,
+    "recall": 0.845541
+   },
+   {
+    "t": 0.2,
+    "precision": 0.059204,
+    "recall": 0.862192
+   },
+   {
+    "t": 0.18,
+    "precision": 0.058856,
+    "recall": 0.874962
+   },
+   {
+    "t": 0.16,
+    "precision": 0.058651,
+    "recall": 0.888855
+   },
+   {
+    "t": 0.14,
+    "precision": 0.058371,
+    "recall": 0.900756
+   },
+   {
+    "t": 0.12,
+    "precision": 0.058127,
+    "recall": 0.913474
+   },
+   {
+    "t": 0.1,
+    "precision": 0.05797,
+    "recall": 0.927725
+   },
+   {
+    "t": 0.08,
+    "precision": 0.057776,
+    "recall": 0.941618
+   },
+   {
+    "t": 0.06,
+    "precision": 0.057623,
+    "recall": 0.955562
+   },
+   {
+    "t": 0.04,
+    "precision": 0.057498,
+    "recall": 0.969609
+   },
+   {
+    "t": 0.02,
+    "precision": 0.057247,
+    "recall": 0.981714
+   },
+   {
+    "t": 0.0,
+    "precision": 0.048634,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w026-ink9um.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:28Z",
+ "scroll": "PHerc0139",
+ "segment": "w026",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w026_ink.png",
+  "sha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791",
+  "width": 629,
+  "height": 640,
+  "pxPerMM": 9.1785,
+  "alignTargetFile": "data/bench/w026_ink.png",
+  "alignTargetSha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791"
+ },
+ "prediction": {
+  "name": "w026_gpu.png",
+  "sha256": "262b1fb5329f95649d5d69da657ac3b111b9ba05c2151859c167efd85c700079",
+  "format": "png",
+  "width": 629,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "scrollprize/ink_9um (seed 42, step 75000) on the 9.362 um scan",
+   "url": "https://huggingface.co/scrollprize/ink_9um"
+  }
+ },
+ "alignment": {
+  "method": "auto",
+  "target": "ink",
+  "tx": 0.531,
+  "ty": 0.969,
+  "sx": 0.99987,
+  "sy": 1.00178,
+  "ncc": 0.29816,
+  "coarseNcc": 0.426747,
+  "evals": 42527,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 401292,
+  "validFraction": 0.99685,
+  "pearson": 0.29816,
+  "blurSigmaPx": 2.0,
+  "blurSigmaUM": 217.9,
+  "pearsonMatchedBlur": 0.324342,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.117089,
+  "dice": 0.209633,
+  "precision": 0.207984,
+  "recall": 0.211309,
+  "refPositiveFraction": 0.048787,
+  "predPositiveFraction": 0.049567,
+  "ap": 0.151895,
+  "bestF1": {
+   "f1": 0.22795,
+   "threshold": 0.32
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.5,
+    "recall": 0.000306
+   },
+   {
+    "t": 0.98,
+    "precision": 0.409463,
+    "recall": 0.022985
+   },
+   {
+    "t": 0.96,
+    "precision": 0.379817,
+    "recall": 0.031719
+   },
+   {
+    "t": 0.94,
+    "precision": 0.374391,
+    "recall": 0.039279
+   },
+   {
+    "t": 0.92,
+    "precision": 0.361761,
+    "recall": 0.046583
+   },
+   {
+    "t": 0.9,
+    "precision": 0.346892,
+    "recall": 0.053581
+   },
+   {
+    "t": 0.88,
+    "precision": 0.337329,
+    "recall": 0.060374
+   },
+   {
+    "t": 0.86,
+    "precision": 0.32753,
+    "recall": 0.067269
+   },
+   {
+    "t": 0.84,
+    "precision": 0.31892,
+    "recall": 0.074216
+   },
+   {
+    "t": 0.82,
+    "precision": 0.311369,
+    "recall": 0.080856
+   },
+   {
+    "t": 0.8,
+    "precision": 0.303041,
+    "recall": 0.087547
+   },
+   {
+    "t": 0.78,
+    "precision": 0.297678,
+    "recall": 0.094954
+   },
+   {
+    "t": 0.76,
+    "precision": 0.290994,
+    "recall": 0.102155
+   },
+   {
+    "t": 0.74,
+    "precision": 0.280074,
+    "recall": 0.107978
+   },
+   {
+    "t": 0.72,
+    "precision": 0.272837,
+    "recall": 0.11564
+   },
+   {
+    "t": 0.7,
+    "precision": 0.267072,
+    "recall": 0.123251
+   },
+   {
+    "t": 0.68,
+    "precision": 0.258982,
+    "recall": 0.130708
+   },
+   {
+    "t": 0.66,
+    "precision": 0.252205,
+    "recall": 0.138778
+   },
+   {
+    "t": 0.64,
+    "precision": 0.242978,
+    "recall": 0.145367
+   },
+   {
+    "t": 0.62,
+    "precision": 0.235898,
+    "recall": 0.153795
+   },
+   {
+    "t": 0.6,
+    "precision": 0.23039,
+    "recall": 0.162172
+   },
+   {
+    "t": 0.58,
+    "precision": 0.224914,
+    "recall": 0.170702
+   },
+   {
+    "t": 0.56,
+    "precision": 0.221309,
+    "recall": 0.180253
+   },
+   {
+    "t": 0.54,
+    "precision": 0.216931,
+    "recall": 0.190571
+   },
+   {
+    "t": 0.52,
+    "precision": 0.212882,
+    "recall": 0.200889
+   },
+   {
+    "t": 0.5,
+    "precision": 0.207984,
+    "recall": 0.211309
+   },
+   {
+    "t": 0.48,
+    "precision": 0.203289,
+    "recall": 0.221626
+   },
+   {
+    "t": 0.46,
+    "precision": 0.19894,
+    "recall": 0.233987
+   },
+   {
+    "t": 0.44,
+    "precision": 0.194885,
+    "recall": 0.247931
+   },
+   {
+    "t": 0.42,
+    "precision": 0.190327,
+    "recall": 0.260701
+   },
+   {
+    "t": 0.4,
+    "precision": 0.185865,
+    "recall": 0.274287
+   },
+   {
+    "t": 0.38,
+    "precision": 0.182261,
+    "recall": 0.288845
+   },
+   {
+    "t": 0.36,
+    "precision": 0.17886,
+    "recall": 0.305445
+   },
+   {
+    "t": 0.34,
+    "precision": 0.175594,
+    "recall": 0.323169
+   },
+   {
+    "t": 0.32,
+    "precision": 0.171351,
+    "recall": 0.340382
+   },
+   {
+    "t": 0.3,
+    "precision": 0.166512,
+    "recall": 0.358157
+   },
+   {
+    "t": 0.28,
+    "precision": 0.162272,
+    "recall": 0.377873
+   },
+   {
+    "t": 0.26,
+    "precision": 0.157717,
+    "recall": 0.398764
+   },
+   {
+    "t": 0.24,
+    "precision": 0.153892,
+    "recall": 0.422617
+   },
+   {
+    "t": 0.22,
+    "precision": 0.147826,
+    "recall": 0.444376
+   },
+   {
+    "t": 0.2,
+    "precision": 0.142145,
+    "recall": 0.467617
+   },
+   {
+    "t": 0.18,
+    "precision": 0.136718,
+    "recall": 0.493871
+   },
+   {
+    "t": 0.16,
+    "precision": 0.130124,
+    "recall": 0.517877
+   },
+   {
+    "t": 0.14,
+    "precision": 0.123865,
+    "recall": 0.546276
+   },
+   {
+    "t": 0.12,
+    "precision": 0.117527,
+    "recall": 0.580243
+   },
+   {
+    "t": 0.1,
+    "precision": 0.111137,
+    "recall": 0.619318
+   },
+   {
+    "t": 0.08,
+    "precision": 0.104007,
+    "recall": 0.663449
+   },
+   {
+    "t": 0.06,
+    "precision": 0.095877,
+    "recall": 0.713709
+   },
+   {
+    "t": 0.04,
+    "precision": 0.086453,
+    "recall": 0.776995
+   },
+   {
+    "t": 0.02,
+    "precision": 0.074829,
+    "recall": 0.866074
+   },
+   {
+    "t": 0.0,
+    "precision": 0.048787,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w026-noise-baseline.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:48:30Z",
+ "scroll": "PHerc0139",
+ "segment": "w026",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w026_ink.png",
+  "sha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791",
+  "width": 629,
+  "height": 640,
+  "pxPerMM": 9.1785,
+  "alignTargetFile": "data/bench/w026_ink.png",
+  "alignTargetSha256": "436cb4c1d102dbf7bfca42b3700f1cf268c29e58a4d5a6f90d130feb9da8e791"
+ },
+ "prediction": {
+  "name": "noise-seed26.png",
+  "sha256": "220a8b63d482620f4a98536fc97a971e7f69bebaafff14199d47b186ae43ffc4",
+  "format": "png",
+  "width": 629,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: uniform noise",
+   "notes": "numpy default_rng(26).random, 8 bit"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": -0.002983,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 402560,
+  "validFraction": 1.0,
+  "pearson": -0.002983,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 653.7,
+  "pearsonMatchedBlur": -0.002789,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.045741,
+  "dice": 0.087481,
+  "precision": 0.04801,
+  "recall": 0.491879,
+  "refPositiveFraction": 0.048634,
+  "predPositiveFraction": 0.498271,
+  "ap": 0.047854,
+  "bestF1": {
+   "f1": 0.092756,
+   "threshold": 0.0
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 1.0,
+    "recall": 0.0
+   },
+   {
+    "t": 0.98,
+    "precision": 0.045315,
+    "recall": 0.018133
+   },
+   {
+    "t": 0.96,
+    "precision": 0.046628,
+    "recall": 0.03744
+   },
+   {
+    "t": 0.94,
+    "precision": 0.046986,
+    "recall": 0.056288
+   },
+   {
+    "t": 0.92,
+    "precision": 0.047393,
+    "recall": 0.07585
+   },
+   {
+    "t": 0.9,
+    "precision": 0.047743,
+    "recall": 0.095566
+   },
+   {
+    "t": 0.88,
+    "precision": 0.048053,
+    "recall": 0.115691
+   },
+   {
+    "t": 0.86,
+    "precision": 0.048044,
+    "recall": 0.134692
+   },
+   {
+    "t": 0.84,
+    "precision": 0.047887,
+    "recall": 0.153386
+   },
+   {
+    "t": 0.82,
+    "precision": 0.047932,
+    "recall": 0.172898
+   },
+   {
+    "t": 0.8,
+    "precision": 0.047496,
+    "recall": 0.194555
+   },
+   {
+    "t": 0.78,
+    "precision": 0.047122,
+    "recall": 0.21187
+   },
+   {
+    "t": 0.76,
+    "precision": 0.046874,
+    "recall": 0.229697
+   },
+   {
+    "t": 0.74,
+    "precision": 0.047309,
+    "recall": 0.250536
+   },
+   {
+    "t": 0.72,
+    "precision": 0.047345,
+    "recall": 0.270201
+   },
+   {
+    "t": 0.7,
+    "precision": 0.047221,
+    "recall": 0.288691
+   },
+   {
+    "t": 0.68,
+    "precision": 0.047153,
+    "recall": 0.307437
+   },
+   {
+    "t": 0.66,
+    "precision": 0.047378,
+    "recall": 0.328174
+   },
+   {
+    "t": 0.64,
+    "precision": 0.047424,
+    "recall": 0.34789
+   },
+   {
+    "t": 0.62,
+    "precision": 0.047463,
+    "recall": 0.368117
+   },
+   {
+    "t": 0.6,
+    "precision": 0.047711,
+    "recall": 0.393094
+   },
+   {
+    "t": 0.58,
+    "precision": 0.047698,
+    "recall": 0.411942
+   },
+   {
+    "t": 0.56,
+    "precision": 0.047593,
+    "recall": 0.430381
+   },
+   {
+    "t": 0.54,
+    "precision": 0.047695,
+    "recall": 0.450046
+   },
+   {
+    "t": 0.52,
+    "precision": 0.047728,
+    "recall": 0.469711
+   },
+   {
+    "t": 0.5,
+    "precision": 0.04801,
+    "recall": 0.491879
+   },
+   {
+    "t": 0.48,
+    "precision": 0.048052,
+    "recall": 0.511595
+   },
+   {
+    "t": 0.46,
+    "precision": 0.048076,
+    "recall": 0.53126
+   },
+   {
+    "t": 0.44,
+    "precision": 0.048,
+    "recall": 0.54975
+   },
+   {
+    "t": 0.42,
+    "precision": 0.048023,
+    "recall": 0.569159
+   },
+   {
+    "t": 0.4,
+    "precision": 0.048048,
+    "recall": 0.592859
+   },
+   {
+    "t": 0.38,
+    "precision": 0.047966,
+    "recall": 0.611196
+   },
+   {
+    "t": 0.36,
+    "precision": 0.048075,
+    "recall": 0.63224
+   },
+   {
+    "t": 0.34,
+    "precision": 0.048121,
+    "recall": 0.652518
+   },
+   {
+    "t": 0.32,
+    "precision": 0.048089,
+    "recall": 0.671366
+   },
+   {
+    "t": 0.3,
+    "precision": 0.048168,
+    "recall": 0.692001
+   },
+   {
+    "t": 0.28,
+    "precision": 0.048165,
+    "recall": 0.711104
+   },
+   {
+    "t": 0.26,
+    "precision": 0.048116,
+    "recall": 0.729594
+   },
+   {
+    "t": 0.24,
+    "precision": 0.048122,
+    "recall": 0.748646
+   },
+   {
+    "t": 0.22,
+    "precision": 0.048233,
+    "recall": 0.769588
+   },
+   {
+    "t": 0.2,
+    "precision": 0.0483,
+    "recall": 0.794055
+   },
+   {
+    "t": 0.18,
+    "precision": 0.048363,
+    "recall": 0.814588
+   },
+   {
+    "t": 0.16,
+    "precision": 0.048424,
+    "recall": 0.835325
+   },
+   {
+    "t": 0.14,
+    "precision": 0.048475,
+    "recall": 0.85591
+   },
+   {
+    "t": 0.12,
+    "precision": 0.048506,
+    "recall": 0.875677
+   },
+   {
+    "t": 0.1,
+    "precision": 0.048505,
+    "recall": 0.895342
+   },
+   {
+    "t": 0.08,
+    "precision": 0.048483,
+    "recall": 0.914547
+   },
+   {
+    "t": 0.06,
+    "precision": 0.048563,
+    "recall": 0.935489
+   },
+   {
+    "t": 0.04,
+    "precision": 0.048558,
+    "recall": 0.955052
+   },
+   {
+    "t": 0.02,
+    "precision": 0.048564,
+    "recall": 0.975023
+   },
+   {
+    "t": 0.0,
+    "precision": 0.048634,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w027-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:50:58Z",
+ "scroll": "PHerc0139",
+ "segment": "w027",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w027_ink.png",
+  "sha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2",
+  "width": 638,
+  "height": 640,
+  "pxPerMM": 9.3357,
+  "alignTargetFile": "data/bench/w027_ink.png",
+  "alignTargetSha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2"
+ },
+ "prediction": {
+  "name": "w027_ct.png",
+  "sha256": "bf5b5d9a5d8574c03dba35ac6b502319aa6ebee01032cff01cc5f7d600ea22b6",
+  "format": "png",
+  "width": 638,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.299446,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 408320,
+  "validFraction": 1.0,
+  "pearson": 0.299446,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 642.7,
+  "pearsonMatchedBlur": 0.355799,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.121584,
+  "dice": 0.216807,
+  "precision": 0.134544,
+  "recall": 0.557949,
+  "refPositiveFraction": 0.093674,
+  "predPositiveFraction": 0.388462,
+  "ap": 0.145828,
+  "bestF1": {
+   "f1": 0.218206,
+   "threshold": 0.44
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.355932,
+    "recall": 0.002196
+   },
+   {
+    "t": 0.98,
+    "precision": 0.365217,
+    "recall": 0.00549
+   },
+   {
+    "t": 0.96,
+    "precision": 0.359572,
+    "recall": 0.009673
+   },
+   {
+    "t": 0.94,
+    "precision": 0.336556,
+    "recall": 0.014562
+   },
+   {
+    "t": 0.92,
+    "precision": 0.304087,
+    "recall": 0.019844
+   },
+   {
+    "t": 0.9,
+    "precision": 0.27519,
+    "recall": 0.026563
+   },
+   {
+    "t": 0.88,
+    "precision": 0.262881,
+    "recall": 0.03655
+   },
+   {
+    "t": 0.86,
+    "precision": 0.237259,
+    "recall": 0.04625
+   },
+   {
+    "t": 0.84,
+    "precision": 0.218432,
+    "recall": 0.05893
+   },
+   {
+    "t": 0.82,
+    "precision": 0.202996,
+    "recall": 0.074041
+   },
+   {
+    "t": 0.8,
+    "precision": 0.188728,
+    "recall": 0.095166
+   },
+   {
+    "t": 0.78,
+    "precision": 0.178574,
+    "recall": 0.11501
+   },
+   {
+    "t": 0.76,
+    "precision": 0.168964,
+    "recall": 0.136213
+   },
+   {
+    "t": 0.74,
+    "precision": 0.162469,
+    "recall": 0.161076
+   },
+   {
+    "t": 0.72,
+    "precision": 0.157527,
+    "recall": 0.188136
+   },
+   {
+    "t": 0.7,
+    "precision": 0.153249,
+    "recall": 0.217339
+   },
+   {
+    "t": 0.68,
+    "precision": 0.150783,
+    "recall": 0.249837
+   },
+   {
+    "t": 0.66,
+    "precision": 0.148056,
+    "recall": 0.282596
+   },
+   {
+    "t": 0.64,
+    "precision": 0.146126,
+    "recall": 0.316531
+   },
+   {
+    "t": 0.62,
+    "precision": 0.144255,
+    "recall": 0.350833
+   },
+   {
+    "t": 0.6,
+    "precision": 0.142247,
+    "recall": 0.392533
+   },
+   {
+    "t": 0.58,
+    "precision": 0.140191,
+    "recall": 0.42605
+   },
+   {
+    "t": 0.56,
+    "precision": 0.138673,
+    "recall": 0.45962
+   },
+   {
+    "t": 0.54,
+    "precision": 0.137246,
+    "recall": 0.493111
+   },
+   {
+    "t": 0.52,
+    "precision": 0.135788,
+    "recall": 0.525661
+   },
+   {
+    "t": 0.5,
+    "precision": 0.134544,
+    "recall": 0.557949
+   },
+   {
+    "t": 0.48,
+    "precision": 0.133308,
+    "recall": 0.589113
+   },
+   {
+    "t": 0.46,
+    "precision": 0.132051,
+    "recall": 0.618578
+   },
+   {
+    "t": 0.44,
+    "precision": 0.13119,
+    "recall": 0.648043
+   },
+   {
+    "t": 0.42,
+    "precision": 0.129787,
+    "recall": 0.674161
+   },
+   {
+    "t": 0.4,
+    "precision": 0.128471,
+    "recall": 0.705666
+   },
+   {
+    "t": 0.38,
+    "precision": 0.127202,
+    "recall": 0.728673
+   },
+   {
+    "t": 0.36,
+    "precision": 0.126083,
+    "recall": 0.751627
+   },
+   {
+    "t": 0.34,
+    "precision": 0.125017,
+    "recall": 0.772569
+   },
+   {
+    "t": 0.32,
+    "precision": 0.123818,
+    "recall": 0.791576
+   },
+   {
+    "t": 0.3,
+    "precision": 0.122866,
+    "recall": 0.810557
+   },
+   {
+    "t": 0.28,
+    "precision": 0.121995,
+    "recall": 0.827891
+   },
+   {
+    "t": 0.26,
+    "precision": 0.12106,
+    "recall": 0.843368
+   },
+   {
+    "t": 0.24,
+    "precision": 0.120205,
+    "recall": 0.857722
+   },
+   {
+    "t": 0.22,
+    "precision": 0.119499,
+    "recall": 0.87231
+   },
+   {
+    "t": 0.2,
+    "precision": 0.118561,
+    "recall": 0.886977
+   },
+   {
+    "t": 0.18,
+    "precision": 0.117871,
+    "recall": 0.899004
+   },
+   {
+    "t": 0.16,
+    "precision": 0.117224,
+    "recall": 0.910246
+   },
+   {
+    "t": 0.14,
+    "precision": 0.116586,
+    "recall": 0.921253
+   },
+   {
+    "t": 0.12,
+    "precision": 0.115769,
+    "recall": 0.930377
+   },
+   {
+    "t": 0.1,
+    "precision": 0.115116,
+    "recall": 0.941018
+   },
+   {
+    "t": 0.08,
+    "precision": 0.114402,
+    "recall": 0.951319
+   },
+   {
+    "t": 0.06,
+    "precision": 0.113659,
+    "recall": 0.961149
+   },
+   {
+    "t": 0.04,
+    "precision": 0.112997,
+    "recall": 0.971084
+   },
+   {
+    "t": 0.02,
+    "precision": 0.112282,
+    "recall": 0.981019
+   },
+   {
+    "t": 0.0,
+    "precision": 0.093674,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w027-ink9um.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:50:57Z",
+ "scroll": "PHerc0139",
+ "segment": "w027",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w027_ink.png",
+  "sha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2",
+  "width": 638,
+  "height": 640,
+  "pxPerMM": 9.3357,
+  "alignTargetFile": "data/bench/w027_ink.png",
+  "alignTargetSha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2"
+ },
+ "prediction": {
+  "name": "w027_gpu.png",
+  "sha256": "89395044a5bf932b7adaeb8522e2d36c87e684f852b6c8ba2e0d30509376e650",
+  "format": "png",
+  "width": 638,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "scrollprize/ink_9um (seed 42, step 75000) on the 9.362 um scan",
+   "url": "https://huggingface.co/scrollprize/ink_9um"
+  }
+ },
+ "alignment": {
+  "method": "auto",
+  "target": "ink",
+  "tx": 1.0,
+  "ty": 0.75,
+  "sx": 1.00556,
+  "sy": 1.00236,
+  "ncc": 0.49414,
+  "coarseNcc": 0.702908,
+  "evals": 42312,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 408320,
+  "validFraction": 1.0,
+  "pearson": 0.49414,
+  "blurSigmaPx": 3.0,
+  "blurSigmaUM": 321.3,
+  "pearsonMatchedBlur": 0.535652,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.266987,
+  "dice": 0.421452,
+  "precision": 0.442537,
+  "recall": 0.402285,
+  "refPositiveFraction": 0.093674,
+  "predPositiveFraction": 0.085154,
+  "ap": 0.383928,
+  "bestF1": {
+   "f1": 0.433898,
+   "threshold": 0.38
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.839416,
+    "recall": 0.003007
+   },
+   {
+    "t": 0.98,
+    "precision": 0.671879,
+    "recall": 0.066276
+   },
+   {
+    "t": 0.96,
+    "precision": 0.653496,
+    "recall": 0.088211
+   },
+   {
+    "t": 0.94,
+    "precision": 0.640202,
+    "recall": 0.105833
+   },
+   {
+    "t": 0.92,
+    "precision": 0.626016,
+    "recall": 0.120787
+   },
+   {
+    "t": 0.9,
+    "precision": 0.614154,
+    "recall": 0.134539
+   },
+   {
+    "t": 0.88,
+    "precision": 0.602493,
+    "recall": 0.147847
+   },
+   {
+    "t": 0.86,
+    "precision": 0.590695,
+    "recall": 0.160318
+   },
+   {
+    "t": 0.84,
+    "precision": 0.582909,
+    "recall": 0.173704
+   },
+   {
+    "t": 0.82,
+    "precision": 0.573791,
+    "recall": 0.187743
+   },
+   {
+    "t": 0.8,
+    "precision": 0.567528,
+    "recall": 0.201051
+   },
+   {
+    "t": 0.78,
+    "precision": 0.559718,
+    "recall": 0.214045
+   },
+   {
+    "t": 0.76,
+    "precision": 0.551387,
+    "recall": 0.227091
+   },
+   {
+    "t": 0.74,
+    "precision": 0.54232,
+    "recall": 0.239719
+   },
+   {
+    "t": 0.72,
+    "precision": 0.533201,
+    "recall": 0.252765
+   },
+   {
+    "t": 0.7,
+    "precision": 0.525344,
+    "recall": 0.265549
+   },
+   {
+    "t": 0.68,
+    "precision": 0.518348,
+    "recall": 0.279929
+   },
+   {
+    "t": 0.66,
+    "precision": 0.51035,
+    "recall": 0.293289
+   },
+   {
+    "t": 0.64,
+    "precision": 0.502141,
+    "recall": 0.306544
+   },
+   {
+    "t": 0.62,
+    "precision": 0.492882,
+    "recall": 0.319538
+   },
+   {
+    "t": 0.6,
+    "precision": 0.484261,
+    "recall": 0.333028
+   },
+   {
+    "t": 0.58,
+    "precision": 0.476978,
+    "recall": 0.347486
+   },
+   {
+    "t": 0.56,
+    "precision": 0.468635,
+    "recall": 0.361526
+   },
+   {
+    "t": 0.54,
+    "precision": 0.460308,
+    "recall": 0.375513
+   },
+   {
+    "t": 0.52,
+    "precision": 0.451121,
+    "recall": 0.388245
+   },
+   {
+    "t": 0.5,
+    "precision": 0.442537,
+    "recall": 0.402285
+   },
+   {
+    "t": 0.48,
+    "precision": 0.433741,
+    "recall": 0.416481
+   },
+   {
+    "t": 0.46,
+    "precision": 0.424769,
+    "recall": 0.43141
+   },
+   {
+    "t": 0.44,
+    "precision": 0.416711,
+    "recall": 0.447489
+   },
+   {
+    "t": 0.42,
+    "precision": 0.40637,
+    "recall": 0.462679
+   },
+   {
+    "t": 0.4,
+    "precision": 0.396898,
+    "recall": 0.477712
+   },
+   {
+    "t": 0.38,
+    "precision": 0.386946,
+    "recall": 0.493817
+   },
+   {
+    "t": 0.36,
+    "precision": 0.377051,
+    "recall": 0.509373
+   },
+   {
+    "t": 0.34,
+    "precision": 0.367086,
+    "recall": 0.526602
+   },
+   {
+    "t": 0.32,
+    "precision": 0.357072,
+    "recall": 0.544406
+   },
+   {
+    "t": 0.3,
+    "precision": 0.346406,
+    "recall": 0.561636
+   },
+   {
+    "t": 0.28,
+    "precision": 0.336663,
+    "recall": 0.581584
+   },
+   {
+    "t": 0.26,
+    "precision": 0.325605,
+    "recall": 0.599963
+   },
+   {
+    "t": 0.24,
+    "precision": 0.314132,
+    "recall": 0.620199
+   },
+   {
+    "t": 0.22,
+    "precision": 0.30291,
+    "recall": 0.641219
+   },
+   {
+    "t": 0.2,
+    "precision": 0.291763,
+    "recall": 0.663024
+   },
+   {
+    "t": 0.18,
+    "precision": 0.279066,
+    "recall": 0.684201
+   },
+   {
+    "t": 0.16,
+    "precision": 0.266522,
+    "recall": 0.706424
+   },
+   {
+    "t": 0.14,
+    "precision": 0.253303,
+    "recall": 0.729875
+   },
+   {
+    "t": 0.12,
+    "precision": 0.239785,
+    "recall": 0.754085
+   },
+   {
+    "t": 0.1,
+    "precision": 0.225313,
+    "recall": 0.77921
+   },
+   {
+    "t": 0.08,
+    "precision": 0.210039,
+    "recall": 0.807341
+   },
+   {
+    "t": 0.06,
+    "precision": 0.193162,
+    "recall": 0.84044
+   },
+   {
+    "t": 0.04,
+    "precision": 0.173651,
+    "recall": 0.879422
+   },
+   {
+    "t": 0.02,
+    "precision": 0.148893,
+    "recall": 0.924259
+   },
+   {
+    "t": 0.0,
+    "precision": 0.093674,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w027-noise-baseline.json

@@ -0,0 +1,335 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:51:00Z",
+ "scroll": "PHerc0139",
+ "segment": "w027",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w027_ink.png",
+  "sha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2",
+  "width": 638,
+  "height": 640,
+  "pxPerMM": 9.3357,
+  "alignTargetFile": "data/bench/w027_ink.png",
+  "alignTargetSha256": "0eda2b76e7922df9d8474b1cbfa09e83ed5d303d474cd0c0dd790fa7b0b820c2"
+ },
+ "prediction": {
+  "name": "noise-seed27.png",
+  "sha256": "3f35963811cc3ba350851a574a55123ebed8a5c217434dc46b650e037f3f1f17",
+  "format": "png",
+  "width": 638,
+  "height": 640,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: uniform noise",
+   "notes": "numpy default_rng(27).random, 8 bit"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": -0.002544,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 408320,
+  "validFraction": 1.0,
+  "pearson": -0.002544,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 642.7,
+  "pearsonMatchedBlur": -0.002194,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.08474,
+  "dice": 0.15624,
+  "precision": 0.092732,
+  "recall": 0.495778,
+  "refPositiveFraction": 0.093674,
+  "predPositiveFraction": 0.500816,
+  "ap": 0.093275,
+  "bestF1": {
+   "f1": 0.171302,
+   "threshold": 0.0
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 1.0,
+    "recall": 0.0
+   },
+   {
+    "t": 0.98,
+    "precision": 0.094288,
+    "recall": 0.019896
+   },
+   {
+    "t": 0.96,
+    "precision": 0.094086,
+    "recall": 0.039557
+   },
+   {
+    "t": 0.94,
+    "precision": 0.093964,
+    "recall": 0.059139
+   },
+   {
+    "t": 0.92,
+    "precision": 0.092801,
+    "recall": 0.078329
+   },
+   {
+    "t": 0.9,
+    "precision": 0.09358,
+    "recall": 0.098695
+   },
+   {
+    "t": 0.88,
+    "precision": 0.093374,
+    "recall": 0.118199
+   },
+   {
+    "t": 0.86,
+    "precision": 0.093263,
+    "recall": 0.137651
+   },
+   {
+    "t": 0.84,
+    "precision": 0.09294,
+    "recall": 0.15671
+   },
+   {
+    "t": 0.82,
+    "precision": 0.093395,
+    "recall": 0.177181
+   },
+   {
+    "t": 0.8,
+    "precision": 0.093183,
+    "recall": 0.200345
+   },
+   {
+    "t": 0.78,
+    "precision": 0.093003,
+    "recall": 0.21964
+   },
+   {
+    "t": 0.76,
+    "precision": 0.093011,
+    "recall": 0.239196
+   },
+   {
+    "t": 0.74,
+    "precision": 0.092967,
+    "recall": 0.2587
+   },
+   {
+    "t": 0.72,
+    "precision": 0.092936,
+    "recall": 0.278282
+   },
+   {
+    "t": 0.7,
+    "precision": 0.092995,
+    "recall": 0.298282
+   },
+   {
+    "t": 0.68,
+    "precision": 0.092958,
+    "recall": 0.317682
+   },
+   {
+    "t": 0.66,
+    "precision": 0.092902,
+    "recall": 0.337211
+   },
+   {
+    "t": 0.64,
+    "precision": 0.092796,
+    "recall": 0.356558
+   },
+   {
+    "t": 0.62,
+    "precision": 0.093038,
+    "recall": 0.377056
+   },
+   {
+    "t": 0.6,
+    "precision": 0.092839,
+    "recall": 0.399514
+   },
+   {
+    "t": 0.58,
+    "precision": 0.092667,
+    "recall": 0.418076
+   },
+   {
+    "t": 0.56,
+    "precision": 0.092578,
+    "recall": 0.436848
+   },
+   {
+    "t": 0.54,
+    "precision": 0.092709,
+    "recall": 0.45677
+   },
+   {
+    "t": 0.52,
+    "precision": 0.092773,
+    "recall": 0.476692
+   },
+   {
+    "t": 0.5,
+    "precision": 0.092732,
+    "recall": 0.495778
+   },
+   {
+    "t": 0.48,
+    "precision": 0.092931,
+    "recall": 0.515987
+   },
+   {
+    "t": 0.46,
+    "precision": 0.093119,
+    "recall": 0.536511
+   },
+   {
+    "t": 0.44,
+    "precision": 0.093081,
+    "recall": 0.556198
+   },
+   {
+    "t": 0.42,
+    "precision": 0.093068,
+    "recall": 0.575335
+   },
+   {
+    "t": 0.4,
+    "precision": 0.093103,
+    "recall": 0.599074
+   },
+   {
+    "t": 0.38,
+    "precision": 0.093244,
+    "recall": 0.619859
+   },
+   {
+    "t": 0.36,
+    "precision": 0.093374,
+    "recall": 0.639625
+   },
+   {
+    "t": 0.34,
+    "precision": 0.093282,
+    "recall": 0.658449
+   },
+   {
+    "t": 0.32,
+    "precision": 0.093285,
+    "recall": 0.677952
+   },
+   {
+    "t": 0.3,
+    "precision": 0.09349,
+    "recall": 0.698868
+   },
+   {
+    "t": 0.28,
+    "precision": 0.093568,
+    "recall": 0.718633
+   },
+   {
+    "t": 0.26,
+    "precision": 0.093552,
+    "recall": 0.737745
+   },
+   {
+    "t": 0.24,
+    "precision": 0.093576,
+    "recall": 0.757039
+   },
+   {
+    "t": 0.22,
+    "precision": 0.093437,
+    "recall": 0.775733
+   },
+   {
+    "t": 0.2,
+    "precision": 0.093393,
+    "recall": 0.798792
+   },
+   {
+    "t": 0.18,
+    "precision": 0.093454,
+    "recall": 0.818793
+   },
+   {
+    "t": 0.16,
+    "precision": 0.093603,
+    "recall": 0.839813
+   },
+   {
+    "t": 0.14,
+    "precision": 0.093634,
+    "recall": 0.859787
+   },
+   {
+    "t": 0.12,
+    "precision": 0.093685,
+    "recall": 0.879683
+   },
+   {
+    "t": 0.1,
+    "precision": 0.093637,
+    "recall": 0.898821
+   },
+   {
+    "t": 0.08,
+    "precision": 0.093631,
+    "recall": 0.918063
+   },
+   {
+    "t": 0.06,
+    "precision": 0.093711,
+    "recall": 0.938142
+   },
+   {
+    "t": 0.04,
+    "precision": 0.093628,
+    "recall": 0.956835
+   },
+   {
+    "t": 0.02,
+    "precision": 0.093714,
+    "recall": 0.97681
+   },
+   {
+    "t": 0.0,
+    "precision": 0.093674,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}

docs/examples/bench/w028-ct-baseline.json

@@ -0,0 +1,334 @@
+{
+ "schema": "unroll-inkbench/1",
+ "tool": {
+  "name": "inkbench",
+  "version": "1.0.0",
+  "impl": "python"
+ },
+ "created": "2026-10-16T03:51:10Z",
+ "scroll": "PHerc0139",
+ "segment": "w028",
+ "reference": {
+  "layer": "ink",
+  "label": "Published ink (2.4 um), Vesuvius Challenge",
+  "file": "data/bench/w028_ink.png",
+  "sha256": "fb51ea9e23deda1cb6583280fb7a2d700979cd5f6f61c6e1dc3652339976a380",
+  "width": 640,
+  "height": 631,
+  "pxPerMM": 9.2888,
+  "alignTargetFile": "data/bench/w028_ink.png",
+  "alignTargetSha256": "fb51ea9e23deda1cb6583280fb7a2d700979cd5f6f61c6e1dc3652339976a380"
+ },
+ "prediction": {
+  "name": "w028_ct.png",
+  "sha256": "02e88630f83b2fea27ec4473e8bf04d43a5c771d8b62e8900fd8f56dc4cc3004",
+  "format": "png",
+  "width": 640,
+  "height": 631,
+  "normalization": "max255",
+  "invert": false,
+  "model": {
+   "name": "baseline: X-ray intensity (same grid, no alignment needed)"
+  }
+ },
+ "alignment": {
+  "method": "manual",
+  "target": "ink",
+  "tx": 0.0,
+  "ty": 0.0,
+  "sx": 1.0,
+  "sy": 1.0,
+  "ncc": 0.29201,
+  "coarseNcc": null,
+  "evals": 0,
+  "model": "bilinear, scale about grid centre, then shift; units are reference px"
+ },
+ "metrics": {
+  "nValid": 403840,
+  "validFraction": 1.0,
+  "pearson": 0.29201,
+  "blurSigmaPx": 6.0,
+  "blurSigmaUM": 645.9,
+  "pearsonMatchedBlur": 0.418986,
+  "thresholds": {
+   "ref": 0.5,
+   "pred": 0.5
+  },
+  "iou": 0.097583,
+  "dice": 0.177814,
+  "precision": 0.105548,
+  "recall": 0.563921,
+  "refPositiveFraction": 0.077419,
+  "predPositiveFraction": 0.413637,
+  "ap": 0.11115,
+  "bestF1": {
+   "f1": 0.177941,
+   "threshold": 0.48
+  },
+  "pr": [
+   {
+    "t": 1.0,
+    "precision": 0.159794,
+    "recall": 0.000992
+   },
+   {
+    "t": 0.98,
+    "precision": 0.20075,
+    "recall": 0.003422
+   },
+   {
+    "t": 0.96,
+    "precision": 0.174542,
+    "recall": 0.005789
+   },
+   {
+    "t": 0.94,
+    "precision": 0.168952,
+    "recall": 0.009851
+   },
+   {
+    "t": 0.92,
+    "precision": 0.163784,
+    "recall": 0.015449
+   },
+   {
+    "t": 0.9,
+    "precision": 0.158648,
+    "recall": 0.022677
+   },
+   {
+    "t": 0.88,
+    "precision": 0.153033,
+    "recall": 0.031953
+   },
+   {
+    "t": 0.86,
+    "precision": 0.149486,
+    "recall": 0.044619
+   },
+   {
+    "t": 0.84,
+    "precision": 0.145116,
+    "recall": 0.059587
+   },
+   {
+    "t": 0.82,
+    "precision": 0.141404,
+    "recall": 0.076923
+   },
+   {
+    "t": 0.8,
+    "precision": 0.137073,
+    "recall": 0.101359
+   },
+   {
+    "t": 0.78,
+    "precision": 0.132283,
+    "recall": 0.123589
+   },
+   {
+    "t": 0.76,
+    "precision": 0.128693,
+    "recall": 0.148089
+   },
+   {
+    "t": 0.74,
+    "precision": 0.126437,
+    "recall": 0.175884
+   },
+   {
+    "t": 0.72,
+    "precision": 0.123469,
+    "recall": 0.204702
+   },
+   {
+    "t": 0.7,
+    "precision": 0.121475,
+    "recall": 0.235887
+   },
+   {
+    "t": 0.68,
+    "precision": 0.118464,
+    "recall": 0.265377
+   },
+   {
+    "t": 0.66,
+    "precision": 0.116752,
+    "recall": 0.298065
+   },
+   {
+    "t": 0.64,
+    "precision": 0.11447,
+    "recall": 0.329538
+   },
+   {
+    "t": 0.62,
+    "precision": 0.112809,
+    "recall": 0.36197
+   },
+   {
+    "t": 0.6,
+    "precision": 0.110745,
+    "recall": 0.400448
+   },
+   {
+    "t": 0.58,
+    "precision": 0.109919,
+    "recall": 0.435343
+   },
+   {
+    "t": 0.56,
+    "precision": 0.108943,
+    "recall": 0.468959
+   },
+   {
+    "t": 0.54,
+    "precision": 0.107364,
+    "recall": 0.498992
+   },
+   {
+    "t": 0.52,
+    "precision": 0.106305,
+    "recall": 0.530945
+   },
+   {
+    "t": 0.5,
+    "precision": 0.105548,
+    "recall": 0.563921
+   },
+   {
+    "t": 0.48,
+    "precision": 0.104655,
+    "recall": 0.593667
+   },
+   {
+    "t": 0.46,
+    "precision": 0.103688,
+    "recall": 0.622901
+   },
+   {
+    "t": 0.44,
+    "precision": 0.10299,
+    "recall": 0.650952
+   },
+   {
+    "t": 0.42,
+    "precision": 0.10223,
+    "recall": 0.677595
+   },
+   {
+    "t": 0.4,
+    "precision": 0.101256,
+    "recall": 0.706413
+   },
+   {
+    "t": 0.38,
+    "precision": 0.100472,
+    "recall": 0.729698
+   },
+   {
+    "t": 0.36,
+    "precision": 0.099899,
+    "recall": 0.753174
+   },
+   {
+    "t": 0.34,
+    "precision": 0.099222,
+    "recall": 0.774828
+   },
+   {
+    "t": 0.32,
+    "precision": 0.098499,
+    "recall": 0.793923
+   },
+   {
+    "t": 0.3,
+    "precision": 0.097848,
+    "recall": 0.811099
+   },
+   {
+    "t": 0.28,
+    "precision": 0.09724,
+    "recall": 0.826963
+   },
+   {
+    "t": 0.26,
+    "precision": 0.096755,
+    "recall": 0.842763
+   },
+   {
+    "t": 0.24,
+    "precision": 0.096275,
+    "recall": 0.857157
+   },
+   {
+    "t": 0.22,
+    "precision": 0.095782,
+    "recall": 0.870302
+   },
+   {
+    "t": 0.2,
+    "precision": 0.095437,
+    "recall": 0.88655
+   },
+   {
+    "t": 0.18,
+    "precision": 0.095042,
+    "recall": 0.898225
+   },
+   {
+    "t": 0.16,
+    "precision": 0.094757,
+    "recall": 0.910251
+   },
+   {
+    "t": 0.14,
+    "precision": 0.094484,
+    "recall": 0.921542
+   },
+   {
+    "t": 0.12,
+    "precision": 0.094234,
+    "recall": 0.93312
+   },
+   {
+    "t": 0.1,
+    "precision": 0.093859,
+    "recall": 0.943355
+   },
+   {
+    "t": 0.08,
+    "precision": 0.093421,
+    "recall": 0.952279
+   },
+   {
+    "t": 0.06,
+    "precision": 0.093045,
+    "recall": 0.962226
+   },
+   {
+    "t": 0.04,
+    "precision": 0.09265,
+    "recall": 0.971534
+   },
+   {
+    "t": 0.02,
+    "precision": 0.092188,
+    "recall": 0.980937
+   },
+   {
+    "t": 0.0,
+    "precision": 0.077419,
+    "recall": 1.0
+   }
+  ]
+ },
+ "annotations": [],
+ "credits": {
+  "data": "PHerc. 139 scan, segments and published ink: Vesuvius Challenge open data (scrollprize.org), ESRF; held by the Biblioteca Nazionale di Napoli.",
+  "dataLicense": "CC BY-NC 4.0 (https://creativecommons.org/licenses/by-nc/4.0/)",
+  "code": "Unroll Ink Bench by Chase Hendrick",
+  "codeLicense": "Apache-2.0"
+ },
+ "disclaimer": "Scores measure agreement with another model's output (the Vesuvius Challenge published ink detection), not agreement with ground truth. Letter boxes are candidates for expert review, not readings."
+}