{
  "analysis_date": "2026-11-05",
  "baseline_merge_commit": "cef329a50aa502162386a7fc0ccbc22f0ffd6b2e",
  "rows": [
    {
      "corpus": "spy-legal",
      "documents": 50,
      "exact_f1_delta": 0.029799447604913876,
      "exact_f1_model_only": 0.2676282051282051,
      "exact_f1_with_rules": 0.297427652733119,
      "fully_contained_candidates": 193,
      "fully_contained_final": 191,
      "high_severity_leaks_candidates": 4,
      "high_severity_leaks_final": 6,
      "invalid_offsets": 0,
      "preset": "en-closed",
      "report": "docs/eval/modal/spy-legal-50-0e100240/spy-legal__en-closed-50.json",
      "report_sha256": "c34498e15ab35dac4a2a51474c72a49fd439a6b72db35b7830e0eef0d0061940",
      "selection_sha256": "8923318576cb7efaa919b3e04a121401f7cc36387f66ce6c564d38ed86bc866c",
      "slice": "all"
    },
    {
      "corpus": "spy-legal",
      "documents": 50,
      "exact_f1_delta": 0.004612626160386379,
      "exact_f1_model_only": 0.37748344370860926,
      "exact_f1_with_rules": 0.38209606986899564,
      "fully_contained_candidates": 218,
      "fully_contained_final": 218,
      "high_severity_leaks_candidates": 11,
      "high_severity_leaks_final": 11,
      "invalid_offsets": 1,
      "preset": "multi-boundary",
      "report": "docs/eval/modal/spy-legal-50-0e100240/spy-legal__multi-boundary-50.json",
      "report_sha256": "90badbfa9c7ef48106ff799045139b5b5c4a53d77f768a4696d64df91f3d390c",
      "selection_sha256": "8923318576cb7efaa919b3e04a121401f7cc36387f66ce6c564d38ed86bc866c",
      "slice": "all"
    },
    {
      "corpus": "spy-legal",
      "documents": 50,
      "exact_f1_delta": -0.0019189661873299513,
      "exact_f1_model_only": 0.4778225806451613,
      "exact_f1_with_rules": 0.47590361445783136,
      "fully_contained_candidates": 247,
      "fully_contained_final": 247,
      "high_severity_leaks_candidates": 0,
      "high_severity_leaks_final": 0,
      "invalid_offsets": 0,
      "preset": "multi-encoder",
      "report": "docs/eval/modal/spy-legal-50-0e100240/spy-legal__multi-encoder-50.json",
      "report_sha256": "3a3720984f54143796b610e93616129033251b98da3bd528e11b06553be461d2",
      "selection_sha256": "8923318576cb7efaa919b3e04a121401f7cc36387f66ce6c564d38ed86bc866c",
      "slice": "all"
    },
    {
      "corpus": "spy-legal",
      "documents": 50,
      "exact_f1_delta": 0.005778211406757661,
      "exact_f1_model_only": 0.44541484716157204,
      "exact_f1_with_rules": 0.4511930585683297,
      "fully_contained_candidates": 240,
      "fully_contained_final": 240,
      "high_severity_leaks_candidates": 3,
      "high_severity_leaks_final": 3,
      "invalid_offsets": 0,
      "preset": "multi-open",
      "report": "docs/eval/modal/spy-legal-50-0e100240/spy-legal__multi-open-50.json",
      "report_sha256": "58cf489f851bb93c6370f2786f98369e696078f8ee8704600f57885e3f35f5bd",
      "selection_sha256": "8923318576cb7efaa919b3e04a121401f7cc36387f66ce6c564d38ed86bc866c",
      "slice": "all"
    },
    {
      "corpus": "spy-medical",
      "documents": 50,
      "exact_f1_delta": 0.019519519519519524,
      "exact_f1_model_only": 0.3138138138138138,
      "exact_f1_with_rules": 0.3333333333333333,
      "fully_contained_candidates": 230,
      "fully_contained_final": 226,
      "high_severity_leaks_candidates": 5,
      "high_severity_leaks_final": 6,
      "invalid_offsets": 0,
      "preset": "en-closed",
      "report": "docs/eval/modal/spy-medical-50-e42b5910/spy-medical__en-closed-50.json",
      "report_sha256": "6a90c0e90ecf70b3da80f6b7fb81d091c9b36eb5445d0ec39feb31ff24fc7174",
      "selection_sha256": "a6fbe20649c5291ea6056f1483304a74aed9edddc5ba33cf93547c5b3be4db39",
      "slice": "all"
    },
    {
      "corpus": "spy-medical",
      "documents": 50,
      "exact_f1_delta": -0.0021150723148383177,
      "exact_f1_model_only": 0.4175152749490835,
      "exact_f1_with_rules": 0.4154002026342452,
      "fully_contained_candidates": 242,
      "fully_contained_final": 240,
      "high_severity_leaks_candidates": 25,
      "high_severity_leaks_final": 25,
      "invalid_offsets": 0,
      "preset": "multi-boundary",
      "report": "docs/eval/modal/spy-medical-50-e42b5910/spy-medical__multi-boundary-50.json",
      "report_sha256": "d0d5b1c8c79560ad57be8645e1d879824b13caf596dec783ea1ad117ca87f2a3",
      "selection_sha256": "a6fbe20649c5291ea6056f1483304a74aed9edddc5ba33cf93547c5b3be4db39",
      "slice": "all"
    },
    {
      "corpus": "spy-medical",
      "documents": 50,
      "exact_f1_delta": -0.011379113512365424,
      "exact_f1_model_only": 0.529688972667295,
      "exact_f1_with_rules": 0.5183098591549296,
      "fully_contained_candidates": 295,
      "fully_contained_final": 293,
      "high_severity_leaks_candidates": 10,
      "high_severity_leaks_final": 10,
      "invalid_offsets": 0,
      "preset": "multi-encoder",
      "report": "docs/eval/modal/spy-medical-50-e42b5910/spy-medical__multi-encoder-50.json",
      "report_sha256": "8d286fb8fdb46feb46863ddcf683f5c010b99d481412c13d8e1d76dd19a3473c",
      "selection_sha256": "a6fbe20649c5291ea6056f1483304a74aed9edddc5ba33cf93547c5b3be4db39",
      "slice": "all"
    },
    {
      "corpus": "spy-medical",
      "documents": 50,
      "exact_f1_delta": 0.0,
      "exact_f1_model_only": 0.5110410094637224,
      "exact_f1_with_rules": 0.5110410094637224,
      "fully_contained_candidates": 271,
      "fully_contained_final": 269,
      "high_severity_leaks_candidates": 23,
      "high_severity_leaks_final": 23,
      "invalid_offsets": 1,
      "preset": "multi-open",
      "report": "docs/eval/modal/spy-medical-50-e42b5910/spy-medical__multi-open-50.json",
      "report_sha256": "9eb35613ac02c6917132d4a98f3a8bae72844cdd9294cc3562257137e083b31c",
      "selection_sha256": "a6fbe20649c5291ea6056f1483304a74aed9edddc5ba33cf93547c5b3be4db39",
      "slice": "all"
    },
    {
      "corpus": "gretel-finance-en",
      "documents": 50,
      "exact_f1_delta": 0.025753589415921235,
      "exact_f1_model_only": 0.1553784860557769,
      "exact_f1_with_rules": 0.18113207547169813,
      "fully_contained_candidates": 50,
      "fully_contained_final": 50,
      "high_severity_leaks_candidates": 5,
      "high_severity_leaks_final": 5,
      "invalid_offsets": 0,
      "preset": "en-closed",
      "report": "docs/eval/modal/gretel-finance-en-50-86e9aa29/gretel-finance-en__en-closed-50.json",
      "report_sha256": "34a6f219e84df0c0f3f2230dd7dd2da2b4ca9f67a2e681a786c7d2ab5da0a3a0",
      "selection_sha256": "e0c16f7df0a7e0a3c8f1c7154f80c15d2f520aec4c5fe14db6439a67d426f8ae",
      "slice": "all"
    },
    {
      "corpus": "gretel-finance-en",
      "documents": 50,
      "exact_f1_delta": -0.027941176470588247,
      "exact_f1_model_only": 0.635,
      "exact_f1_with_rules": 0.6070588235294118,
      "fully_contained_candidates": 138,
      "fully_contained_final": 138,
      "high_severity_leaks_candidates": 5,
      "high_severity_leaks_final": 5,
      "invalid_offsets": 0,
      "preset": "multi-boundary",
      "report": "docs/eval/modal/gretel-finance-en-50-86e9aa29/gretel-finance-en__multi-boundary-50.json",
      "report_sha256": "e28a455d067e7112546742dd86ef88c12afab59c2337866a640af33ca6bffb6d",
      "selection_sha256": "e0c16f7df0a7e0a3c8f1c7154f80c15d2f520aec4c5fe14db6439a67d426f8ae",
      "slice": "all"
    },
    {
      "corpus": "gretel-finance-en",
      "documents": 50,
      "exact_f1_delta": 0.0035074958580838533,
      "exact_f1_model_only": 0.6430678466076696,
      "exact_f1_with_rules": 0.6465753424657534,
      "fully_contained_candidates": 126,
      "fully_contained_final": 126,
      "high_severity_leaks_candidates": 4,
      "high_severity_leaks_final": 4,
      "invalid_offsets": 0,
      "preset": "multi-encoder",
      "report": "docs/eval/modal/gretel-finance-en-50-86e9aa29/gretel-finance-en__multi-encoder-50.json",
      "report_sha256": "b691567596f103da2adbd113dd093dc4bfca84b00a8f35eaf5f5d923972361e0",
      "selection_sha256": "e0c16f7df0a7e0a3c8f1c7154f80c15d2f520aec4c5fe14db6439a67d426f8ae",
      "slice": "all"
    },
    {
      "corpus": "gretel-finance-en",
      "documents": 50,
      "exact_f1_delta": -0.022322274881516613,
      "exact_f1_model_only": 0.61,
      "exact_f1_with_rules": 0.5876777251184834,
      "fully_contained_candidates": 136,
      "fully_contained_final": 136,
      "high_severity_leaks_candidates": 0,
      "high_severity_leaks_final": 0,
      "invalid_offsets": 0,
      "preset": "multi-open",
      "report": "docs/eval/modal/gretel-finance-en-50-86e9aa29/gretel-finance-en__multi-open-50.json",
      "report_sha256": "4639a367b284600ee537814b09c44192850ed1681042f25eccd82fdcfb7fa541",
      "selection_sha256": "e0c16f7df0a7e0a3c8f1c7154f80c15d2f520aec4c5fe14db6439a67d426f8ae",
      "slice": "all"
    },
    {
      "corpus": "nemotron-pii",
      "documents": 25,
      "exact_f1_delta": -0.08243534482758619,
      "exact_f1_model_only": 0.8793103448275862,
      "exact_f1_with_rules": 0.796875,
      "fully_contained_candidates": 51,
      "fully_contained_final": 51,
      "high_severity_leaks_candidates": 2,
      "high_severity_leaks_final": 2,
      "invalid_offsets": 0,
      "preset": "en-closed",
      "report": "docs/eval/modal/nemotron-pii-25-a9d52b9f/nemotron-pii__en-closed-25.json",
      "report_sha256": "7eadcb2494c51ba42a63474d25fda91c4519c3b421dc29e44055350a0c9624da",
      "selection_sha256": "56a52feb9d8b4fa14087c5e4be7e6bd5eca609f8fe88ea17678855114fd38b3d",
      "slice": "all"
    },
    {
      "corpus": "nemotron-pii",
      "documents": 25,
      "exact_f1_delta": -0.049026548672566284,
      "exact_f1_model_only": 0.58,
      "exact_f1_with_rules": 0.5309734513274337,
      "fully_contained_candidates": 41,
      "fully_contained_final": 41,
      "high_severity_leaks_candidates": 9,
      "high_severity_leaks_final": 9,
      "invalid_offsets": 0,
      "preset": "multi-boundary",
      "report": "docs/eval/modal/nemotron-pii-25-a9d52b9f/nemotron-pii__multi-boundary-25.json",
      "report_sha256": "e2e27e8f991a5eca4abff6427770838fa88315d74783dc25ad6cbb17249c34b6",
      "selection_sha256": "56a52feb9d8b4fa14087c5e4be7e6bd5eca609f8fe88ea17678855114fd38b3d",
      "slice": "all"
    },
    {
      "corpus": "nemotron-pii",
      "documents": 25,
      "exact_f1_delta": -0.07066073262581318,
      "exact_f1_model_only": 0.7478260869565218,
      "exact_f1_with_rules": 0.6771653543307086,
      "fully_contained_candidates": 51,
      "fully_contained_final": 51,
      "high_severity_leaks_candidates": 2,
      "high_severity_leaks_final": 2,
      "invalid_offsets": 0,
      "preset": "multi-encoder",
      "report": "docs/eval/modal/nemotron-pii-25-a9d52b9f/nemotron-pii__multi-encoder-25.json",
      "report_sha256": "ee9c9149b0328a7f83803ef5bb6b1c152e972441b955c28ec4a1c5865300b1be",
      "selection_sha256": "56a52feb9d8b4fa14087c5e4be7e6bd5eca609f8fe88ea17678855114fd38b3d",
      "slice": "all"
    },
    {
      "corpus": "nemotron-pii",
      "documents": 25,
      "exact_f1_delta": -0.0381231671554253,
      "exact_f1_model_only": 0.6290322580645161,
      "exact_f1_with_rules": 0.5909090909090908,
      "fully_contained_candidates": 48,
      "fully_contained_final": 48,
      "high_severity_leaks_candidates": 2,
      "high_severity_leaks_final": 2,
      "invalid_offsets": 0,
      "preset": "multi-open",
      "report": "docs/eval/modal/nemotron-pii-25-a9d52b9f/nemotron-pii__multi-open-25.json",
      "report_sha256": "1bc022526ac70143b6d7158d02e77b27d74f2caecbda2ef01e8427f8bd9ffb59",
      "selection_sha256": "56a52feb9d8b4fa14087c5e4be7e6bd5eca609f8fe88ea17678855114fd38b3d",
      "slice": "all"
    }
  ],
  "scope": "Reanalysis of existing aggregate measurements; no new inference, no statistical significance claim. Counts use the all prefix slice, not the sum of overlapping slices. min_score was null in these reports.",
  "summary": {
    "paired_rows": 16,
    "policy_added_high_severity_misses": 3,
    "rules_higher_exact_f1": 6,
    "rules_lower_exact_f1": 9,
    "ties": 1
  }
}