{
  "annotationCampaign": {
    "campaignId": "fc-3377c49ce724",
    "coverage": {
      "cellsPending": 1,
      "cellsRun": 14,
      "complete": false,
      "completedAttempts": 136,
      "examsNotStarted": [],
      "missingAttempts": 4,
      "note": "Cells run against the cells this campaign declared it would run.",
      "pending": [
        {
          "completedAttempts": 6,
          "environmentId": "vvdex.annotation.text-labeling-1",
          "missingAttempts": 4,
          "model": "openai/gpt-oss-120b",
          "plannedAttempts": 10
        }
      ],
      "planDeclared": true,
      "plannedAttempts": 140,
      "plannedCells": 14,
      "plannedExams": [
        "vvdex.annotation.text-labeling-1",
        "vvdex.annotation.structured-data-1"
      ],
      "plannedModels": [
        "openai/gpt-oss-20b",
        "openai/gpt-oss-120b",
        "@cf/openai/gpt-oss-20b",
        "codestral-latest",
        "mistral-code-latest",
        "ministral-14b-latest",
        "ministral-8b-latest"
      ],
      "rolloutsPerCell": 10
    },
    "limitations": [
      "Repeated attempts on these fixed tasks only; not a broad capability score.",
      "All outcomes are retained; lane errors, withheld and invalid attempts are not model failures.",
      "Quality is the existing grader measurement, separate from the pass gate.",
      "Video delivery is stated per model from recorded receipts; native files and sampled frames are distinct. Provider-side sampling is not independently observed.",
      "Provider token usage is recorded where available. Monetary usage is not an invoice, and requested seeds do not establish deterministic replay.",
      "Private references, submissions, exact geometry, frame timestamps and raw sessions are withheld."
    ],
    "schema": "vvdex.forge.annotation-model-campaign/v1",
    "tasks": [
      {
        "environmentId": "vvdex.annotation.structured-data-1",
        "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
        "modality": "structured",
        "models": [
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "openai/gpt-oss-20b",
            "passed": 10,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowAccuracy",
                  "mean": 1.0,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 1.0
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 10,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 3,
            "invalid": 0,
            "laneErrors": 0,
            "model": "openai/gpt-oss-120b",
            "passed": 7,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.9,
                  "n": 10
                },
                {
                  "id": "rowRecall",
                  "mean": 0.9,
                  "n": 10
                },
                {
                  "id": "rowF1",
                  "mean": 0.9,
                  "n": 10
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.9,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.925
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 7,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "@cf/openai/gpt-oss-20b",
            "passed": 9,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.9666666666666668,
                  "n": 10
                },
                {
                  "id": "rowRecall",
                  "mean": 0.9666666666666668,
                  "n": 10
                },
                {
                  "id": "rowF1",
                  "mean": 0.9666666666666668,
                  "n": 10
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.9666666666666668,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.975
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 9,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "codestral-latest",
            "passed": 9,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.9666666666666666,
                  "n": 10
                },
                {
                  "id": "rowRecall",
                  "mean": 0.9666666666666666,
                  "n": 10
                },
                {
                  "id": "rowF1",
                  "mean": 0.9666666666666666,
                  "n": 10
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.9666666666666666,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.975
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 9,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "mistral-code-latest",
            "passed": 9,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "rowAccuracy",
                  "mean": 1.0,
                  "n": 9
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.9
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 9,
              "inspect": 10,
              "submit": 9,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 2,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-14b-latest",
            "passed": 8,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 9
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.9629629629629629,
                  "n": 9
                },
                {
                  "id": "rowRecall",
                  "mean": 0.9629629629629629,
                  "n": 9
                },
                {
                  "id": "rowF1",
                  "mean": 0.9629629629629629,
                  "n": 9
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.9629629629629629,
                  "n": 9
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.875
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 8,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 9,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-8b-latest",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.7000000000000001,
                  "n": 10
                },
                {
                  "id": "rowRecall",
                  "mean": 0.7000000000000001,
                  "n": 10
                },
                {
                  "id": "rowF1",
                  "mean": 0.7000000000000001,
                  "n": 10
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.7000000000000001,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.775
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 1,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          }
        ]
      },
      {
        "environmentId": "vvdex.annotation.text-labeling-1",
        "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
        "modality": "text",
        "models": [
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 9,
            "model": "openai/gpt-oss-20b",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "spanPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 1.0,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 1.0
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 1,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 6,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 3,
            "model": "openai/gpt-oss-120b",
            "passed": 3,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "spanPrecision",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "spanRecall",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "spanF1",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 1.0,
                  "n": 3
                },
                {
                  "id": "spanExactF1",
                  "mean": 1.0,
                  "n": 3
                }
              ],
              "measuredAttempts": 3,
              "scoreMean": 1.0
            },
            "stages": {
              "edit": 3,
              "graded": 3,
              "hidden_pass": 3,
              "inspect": 3,
              "submit": 3,
              "test": 3
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "@cf/openai/gpt-oss-20b",
            "passed": 10,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanExactF1",
                  "mean": 1.0,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 1.0
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 10,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 10,
            "invalid": 0,
            "laneErrors": 0,
            "model": "codestral-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 7
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 7
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 7
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 7
                },
                {
                  "id": "spanPrecision",
                  "mean": 0.5,
                  "n": 7
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 7
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 7
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.5873015873015872,
                  "n": 7
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 7
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.5074074074074074
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 0,
              "inspect": 10,
              "submit": 7,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 10,
            "invalid": 0,
            "laneErrors": 0,
            "model": "mistral-code-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 6
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 6
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 6
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 6
                },
                {
                  "id": "spanPrecision",
                  "mean": 0.5833333333333334,
                  "n": 6
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5833333333333334,
                  "n": 6
                },
                {
                  "id": "spanF1",
                  "mean": 0.5833333333333334,
                  "n": 6
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.6597222222222222,
                  "n": 6
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 6
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.4638888888888889
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 0,
              "inspect": 10,
              "submit": 6,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 10,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-14b-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanPrecision",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.6666666666666667
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 0,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          },
          {
            "attempts": 10,
            "delivery": "canonical text tools",
            "failed": 10,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-8b-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "classificationPrecision",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationRecall",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationF1",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "classificationAccuracy",
                  "mean": 1.0,
                  "n": 10
                },
                {
                  "id": "spanPrecision",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 10
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.49000000000000005,
                  "n": 10
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.45,
                  "n": 10
                }
              ],
              "measuredAttempts": 10,
              "scoreMean": 0.6599999999999999
            },
            "stages": {
              "edit": 10,
              "graded": 10,
              "hidden_pass": 0,
              "inspect": 10,
              "submit": 10,
              "test": 10
            },
            "withheld": 0
          }
        ]
      }
    ]
  },
  "campaignId": "fc-3377c49ce724",
  "counts": {
    "attempts": 136,
    "cells": 14,
    "certifiedPasses": 67,
    "exams": 2,
    "graded": 124,
    "invalid": 0,
    "laneErrors": 12,
    "lanes": 7,
    "modelFails": 57,
    "withheld": 0
  },
  "disclosure": {
    "vvdex.annotation.structured-data-1": {
      "class": "answer_key",
      "rule": "answer-key disclosure (ownership vvdex_proprietary \u2014 no declaration \u2014 proprietary by default; family annotation): the exam is VVDex evaluation IP and the graded output is an answer or a keyed value, so no submission content, diff, model claim or hidden-assertion name is published."
    },
    "vvdex.annotation.text-labeling-1": {
      "class": "answer_key",
      "rule": "answer-key disclosure (ownership vvdex_proprietary \u2014 no declaration \u2014 proprietary by default; family annotation): the exam is VVDex evaluation IP and the graded output is an answer or a keyed value, so no submission content, diff, model claim or hidden-assertion name is published."
    }
  },
  "families": [
    {
      "envId": "vvdex.annotation.structured-data-1",
      "family": "annotation",
      "label": "Multimodal annotation",
      "short": "Annotation \u00b7 structured-data-1",
      "what": "Invented tabular records are reviewed for normalization, validity, duplicates, anomalies and relationships."
    },
    {
      "envId": "vvdex.annotation.text-labeling-1",
      "family": "annotation",
      "label": "Multimodal annotation",
      "short": "Annotation \u00b7 text-labeling-1",
      "what": "An invented message is reviewed for classification and entity spans with consistent offsets."
    }
  ],
  "lanes": [
    {
      "certifiedPasses": 19,
      "countable": 20,
      "counts": {
        "certified_pass": 19,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 1,
        "withheld": 0
      },
      "lane": "@cf/openai/gpt-oss-20b",
      "rate": 0.95,
      "rateInterval95": [
        0.7639,
        0.9911
      ]
    },
    {
      "certifiedPasses": 11,
      "countable": 11,
      "counts": {
        "certified_pass": 11,
        "invalid": 0,
        "lane_error": 9,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "openai/gpt-oss-20b",
      "rate": 1.0,
      "rateInterval95": [
        0.7412,
        1.0
      ]
    },
    {
      "certifiedPasses": 10,
      "countable": 13,
      "counts": {
        "certified_pass": 10,
        "invalid": 0,
        "lane_error": 3,
        "model_fail": 3,
        "withheld": 0
      },
      "lane": "openai/gpt-oss-120b",
      "rate": 0.7692307692307693,
      "rateInterval95": [
        0.4974,
        0.9182
      ]
    },
    {
      "certifiedPasses": 9,
      "countable": 20,
      "counts": {
        "certified_pass": 9,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 11,
        "withheld": 0
      },
      "lane": "codestral-latest",
      "rate": 0.45,
      "rateInterval95": [
        0.2582,
        0.6579
      ]
    },
    {
      "certifiedPasses": 9,
      "countable": 20,
      "counts": {
        "certified_pass": 9,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 11,
        "withheld": 0
      },
      "lane": "mistral-code-latest",
      "rate": 0.45,
      "rateInterval95": [
        0.2582,
        0.6579
      ]
    },
    {
      "certifiedPasses": 8,
      "countable": 20,
      "counts": {
        "certified_pass": 8,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 12,
        "withheld": 0
      },
      "lane": "ministral-14b-latest",
      "rate": 0.4,
      "rateInterval95": [
        0.2188,
        0.6134
      ]
    },
    {
      "certifiedPasses": 1,
      "countable": 20,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 19,
        "withheld": 0
      },
      "lane": "ministral-8b-latest",
      "rate": 0.05,
      "rateInterval95": [
        0.0089,
        0.2361
      ]
    }
  ],
  "perExamMatrix": {
    "exams": [
      {
        "envId": "vvdex.annotation.structured-data-1",
        "evalId": "fr-20260907-811ca1e7",
        "family": "annotation",
        "short": "Annotation \u00b7 structured-data-1",
        "title": "Record normalization and duplicate QA"
      },
      {
        "envId": "vvdex.annotation.text-labeling-1",
        "evalId": "fr-20260907-d7a267e3",
        "family": "annotation",
        "short": "Annotation \u00b7 text-labeling-1",
        "title": "Intent and entity annotation"
      }
    ],
    "rows": [
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 9,
            "text": "9/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 10,
            "text": "10/10",
            "withheld": 0
          }
        },
        "certifiedPasses": 19,
        "lane": "@cf/openai/gpt-oss-20b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 10,
            "text": "10/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 9,
            "passes": 1,
            "text": "1/1 + 9 lane errors",
            "withheld": 0
          }
        },
        "certifiedPasses": 11,
        "lane": "openai/gpt-oss-20b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 7,
            "text": "7/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 3,
            "invalid": 0,
            "laneErrors": 3,
            "passes": 3,
            "text": "3/3 + 3 lane errors",
            "withheld": 0
          }
        },
        "certifiedPasses": 10,
        "lane": "openai/gpt-oss-120b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 9,
            "text": "9/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/10",
            "withheld": 0
          }
        },
        "certifiedPasses": 9,
        "lane": "codestral-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 9,
            "text": "9/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/10",
            "withheld": 0
          }
        },
        "certifiedPasses": 9,
        "lane": "mistral-code-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 8,
            "text": "8/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/10",
            "withheld": 0
          }
        },
        "certifiedPasses": 8,
        "lane": "ministral-14b-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/10",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 10,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/10",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "ministral-8b-latest"
      }
    ]
  },
  "recordCount": 4,
  "records": [
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.structured-data-1",
      "evalId": "fr-20260907-811ca1e7",
      "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
      "recordDigest": "50cbe9a820d64927b58510039c580bbed3bd161565d085363f841eb867181fab",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.structured-data-1",
      "evalId": "fr-20260907-aa28bed9",
      "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
      "recordDigest": "b5bf2458f67d7fef214fedfc66497d235bb436dfd5ca1a13080b1eb7f14e00dd",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.text-labeling-1",
      "evalId": "fr-20260907-d7a267e3",
      "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
      "recordDigest": "7286bf674cfcf3482e06f76daaf1f68a1db1cbf89725833b7c3eb0f9f76e8009",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.text-labeling-1",
      "evalId": "fr-20260907-dad35536",
      "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
      "recordDigest": "d98a068bbbb888bb6215a3e1a72a7d079fc043fd8b98f52d14be0a58303da22e",
      "retired_for_future_evaluation": null
    }
  ],
  "refused": 0,
  "retired_for_future_evaluation": {},
  "schema": "vvdex.forge.campaign-summary/v1",
  "suspectRecords": [
    "vvdex.annotation.text-labeling-1"
  ],
  "withheldByHarness": 0
}