{
  "schema_version": 1,
  "updated": "2026-09-26",
  "updated_note": "Website snapshot preparation date; not model evaluation completion.",
  "tracks": [
    {
      "id": "paper-2026-09-26",
      "label": "Paper snapshot · full release",
      "description": "Results transcribed from the supplied manuscript (26 September 2026). Development and test are evaluated jointly; these are not held-out test estimates.",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "reference": "Manuscript Table 2; Appendix B.3–B.5; Tables 8–11.",
      "reference_url": "",
      "pools": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "P": {
          "parents": 385,
          "probes": 553
        },
        "solve": {
          "parents": 986
        }
      },
      "note": "The snapshot identifier is a website bookkeeping label, not a published dataset tag. Audit-corrected data and supplementary candidates are not substituted into these paper results."
    }
  ],
  "models": [
    {
      "id": "qwen35-4b-paper",
      "name": "Qwen3.5-4B",
      "family": "Qwen",
      "access": "open",
      "parameters_b": 4,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-15",
      "config": {
        "checkpoint_or_alias": "Qwen3.5-4B",
        "reasoning": "Native thinking",
        "sampling": "Greedy",
        "output_cap": 8192,
        "image_policy": "IMG02: min_pixels=65536, max_pixels=262144; patch size=16, merge size=2."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 40.21,
        "cacc": 53.97,
        "jacc": 32.09,
        "gacc_joint": 46.51,
        "solveacc": 9.94
      },
      "paired": {
        "parents": 311,
        "probes": 412,
        "base": 46.51,
        "gt": 54.61,
        "delta_pp": 8.1,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 40.54,
        "joint_quadrants": [
          32.09,
          21.88,
          14.42,
          31.62
        ]
      },
      "notes": "Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    },
    {
      "id": "qwen35-9b-paper",
      "name": "Qwen3.5-9B",
      "family": "Qwen",
      "access": "open",
      "parameters_b": 9,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-14",
      "config": {
        "checkpoint_or_alias": "Qwen3.5-9B",
        "reasoning": "Native thinking",
        "sampling": "Greedy",
        "output_cap": 8192,
        "image_policy": "IMG02: min_pixels=65536, max_pixels=262144; patch size=16, merge size=2."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 40.52,
        "cacc": 65.81,
        "jacc": 37.55,
        "gacc_joint": 54.55,
        "solveacc": 12.12
      },
      "paired": {
        "parents": 311,
        "probes": 412,
        "base": 54.55,
        "gt": 56.53,
        "delta_pp": 1.98,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 42.94,
        "joint_quadrants": [
          37.55,
          28.26,
          17.0,
          17.19
        ]
      },
      "notes": "Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    },
    {
      "id": "qwen35-27b-paper",
      "name": "Qwen3.5-27B",
      "family": "Qwen",
      "access": "open",
      "parameters_b": 27,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-15",
      "config": {
        "checkpoint_or_alias": "Qwen3.5-27B",
        "reasoning": "Native thinking",
        "sampling": "Greedy",
        "output_cap": 8192,
        "image_policy": "IMG02: min_pixels=65536, max_pixels=262144; patch size=16, merge size=2."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 84.11,
        "cacc": 84.14,
        "jacc": 68.12,
        "gacc_joint": 80.29,
        "solveacc": 18.39
      },
      "paired": {
        "parents": 311,
        "probes": 412,
        "base": 80.29,
        "gt": 82.0,
        "delta_pp": 1.71,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 19.03,
        "joint_quadrants": [
          68.12,
          16.01,
          12.17,
          3.7
        ]
      },
      "notes": "Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    },
    {
      "id": "internvl35-8b-paper",
      "name": "InternVL3.5-8B",
      "family": "InternVL",
      "access": "open",
      "parameters_b": 8,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-15",
      "config": {
        "checkpoint_or_alias": "InternVL3.5-8B-HF",
        "reasoning": "Explicit CoT",
        "sampling": "Greedy",
        "output_cap": 8192,
        "image_policy": "IMG01: single-patch input; crop_to_patches=true, min_patches=max_patches=1; saved target 448 × 448."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 23.52,
        "cacc": 2.12,
        "jacc": 1.05,
        "gacc_joint": 33.36,
        "solveacc": 7.47
      },
      "paired": {
        "parents": 311,
        "probes": 412,
        "base": 33.36,
        "gt": 34.81,
        "delta_pp": 1.45,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 50.63,
        "joint_quadrants": [
          1.05,
          1.07,
          32.32,
          65.57
        ]
      },
      "notes": "The low recognition score and strong interface sensitivity are retained exactly as reported; they should not be silently repaired or removed. Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    },
    {
      "id": "gpt6-astra-paper",
      "name": "GPT-6 Astra",
      "family": "GPT",
      "access": "api",
      "parameters_b": null,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-15",
      "config": {
        "checkpoint_or_alias": "gpt-6-astra",
        "reasoning": "High effort",
        "sampling": "Provider default (temperature omitted)",
        "output_cap": 16384,
        "image_policy": "IMG03: provider-managed; detail=auto; saved image bytes in source order without client-side resizing."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 81.94,
        "cacc": 89.44,
        "jacc": 77.13,
        "gacc_joint": 83.31,
        "solveacc": 71.72
      },
      "paired": {
        "parents": 304,
        "probes": 397,
        "base": 83.59,
        "gt": 84.9,
        "delta_pp": 1.32,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 13.77,
        "joint_quadrants": [
          77.13,
          12.32,
          6.19,
          4.37
        ]
      },
      "notes": "Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    },
    {
      "id": "gemini38-flash-paper",
      "name": "Gemini 3.8 Flash",
      "family": "Gemini",
      "access": "api",
      "parameters_b": null,
      "model_url": "",
      "track_id": "paper-2026-09-26",
      "benchmark_id": "paper-retained-3341",
      "protocol_id": "paper-balanced-v1",
      "status": "paper-reported",
      "result_source": "Table 2; Appendix B.5; Tables 8–10.",
      "artifact_url": "",
      "dataset_revision": null,
      "evaluator_commit": null,
      "membership_sha256": null,
      "run_created_utc": "2026-09-15",
      "config": {
        "checkpoint_or_alias": "gemini-3.8-flash",
        "reasoning": "Provider default (reasoning control omitted)",
        "sampling": "Temperature 0",
        "output_cap": 8192,
        "image_policy": "IMG03: provider-managed; detail=auto; saved image bytes in source order without client-side resizing."
      },
      "support": {
        "G": {
          "parents": 986,
          "probes": 3341
        },
        "L": {
          "parents": 311,
          "probes": 412
        },
        "solve": {
          "parents": 986
        }
      },
      "metrics": {
        "gacc_all": 90.71,
        "cacc": 85.2,
        "jacc": 70.59,
        "gacc_joint": 79.33,
        "solveacc": 21.31
      },
      "paired": {
        "parents": 304,
        "probes": 397,
        "base": 78.67,
        "gt": 80.86,
        "delta_pp": 2.2,
        "ci95_pp": null,
        "membership_sha256": null
      },
      "diagnostics": {
        "grounding_error_given_correct_reading": 17.15,
        "joint_quadrants": [
          70.59,
          14.61,
          8.74,
          6.05
        ]
      },
      "notes": "Point estimates are reported under model-specific configurations, not equal-compute budgets. Run date denotes creation, not completion."
    }
  ]
}