{
  "annotationCampaign": {
    "campaignId": "fc-0c7e7c67a231",
    "coverage": {
      "cellsPending": 0,
      "cellsRun": 34,
      "complete": true,
      "completedAttempts": 40,
      "examsNotStarted": [],
      "missingAttempts": 0,
      "note": "Cells run against the cells this campaign declared it would run.",
      "pending": [],
      "planDeclared": true,
      "plannedAttempts": 34,
      "plannedCells": 34,
      "plannedExams": [
        "vvdex.annotation.text-labeling-1",
        "vvdex.annotation.structured-data-1"
      ],
      "plannedModels": [
        "openai/gpt-oss-20b",
        "openai/gpt-oss-120b",
        "mistral-small-latest",
        "codestral-latest",
        "mistral-code-latest",
        "mistral-medium-latest",
        "magistral-small-latest",
        "ministral-14b-latest",
        "ministral-8b-latest",
        "@cf/openai/gpt-oss-20b",
        "nvidia/nemotron-3-super-120b-a12b:free",
        "cohere/north-mini-code:free",
        "minimax/minimax-m3:free",
        "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free",
        "cli/codex",
        "cli/claude-sonnet",
        "cli/cursor-grok"
      ],
      "rolloutsPerCell": 1
    },
    "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cli/codex",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cli/claude-sonnet",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 1,
            "laneErrors": 0,
            "model": "cli/cursor-grok",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "mistral-small-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "mistral-medium-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "magistral-small-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "openai/gpt-oss-20b",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "openai/gpt-oss-120b",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "codestral-latest",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "mistral-code-latest",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-14b-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.6666666666666666,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.75
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-8b-latest",
            "passed": 0,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 0.6666666666666666,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "mean": 0.6666666666666666,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.75
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "@cf/openai/gpt-oss-20b",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "nvidia/nemotron-3-super-120b-a12b:free",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 0,
              "test": 0
            },
            "withheld": 1
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cohere/north-mini-code:free",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 0,
              "test": 0
            },
            "withheld": 1
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "minimax/minimax-m3:free",
            "passed": 1,
            "quality": {
              "axes": [
                {
                  "id": "pairPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "pairAccuracy",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowPrecision",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowRecall",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowF1",
                  "mean": 1.0,
                  "n": 1
                },
                {
                  "id": "rowAccuracy",
                  "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 1,
            "model": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          }
        ]
      },
      {
        "environmentId": "vvdex.annotation.text-labeling-1",
        "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
        "modality": "text",
        "models": [
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cli/codex",
            "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cli/claude-sonnet",
            "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 1,
            "laneErrors": 0,
            "model": "cli/cursor-grok",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "mistral-small-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "mistral-medium-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 2,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 2,
            "model": "magistral-small-latest",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 1,
            "model": "openai/gpt-oss-20b",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "openai/gpt-oss-120b",
            "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": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "codestral-latest",
            "passed": 0,
            "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": 0.5,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.625,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.75
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "mistral-code-latest",
            "passed": 0,
            "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": 0.5,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.5555555555555556,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.7037037037037037
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-14b-latest",
            "passed": 0,
            "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": 0.5,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.6666666666666666
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "ministral-8b-latest",
            "passed": 0,
            "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": 0.5,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.4,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.0,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.6
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "@cf/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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "nvidia/nemotron-3-super-120b-a12b:free",
            "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": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 0,
            "model": "cohere/north-mini-code:free",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 0,
              "test": 0
            },
            "withheld": 1
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 1,
            "invalid": 0,
            "laneErrors": 0,
            "model": "minimax/minimax-m3:free",
            "passed": 0,
            "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": 0.5,
                  "n": 1
                },
                {
                  "id": "spanRecall",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanF1",
                  "mean": 0.5,
                  "n": 1
                },
                {
                  "id": "spanOverlapIoU",
                  "mean": 0.625,
                  "n": 1
                },
                {
                  "id": "spanExactF1",
                  "mean": 0.5,
                  "n": 1
                }
              ],
              "measuredAttempts": 1,
              "scoreMean": 0.75
            },
            "stages": {
              "edit": 1,
              "graded": 1,
              "hidden_pass": 0,
              "inspect": 1,
              "submit": 1,
              "test": 1
            },
            "withheld": 0
          },
          {
            "attempts": 1,
            "delivery": "canonical text tools",
            "failed": 0,
            "invalid": 0,
            "laneErrors": 1,
            "model": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free",
            "passed": 0,
            "quality": null,
            "stages": {
              "edit": 0,
              "graded": 0,
              "hidden_pass": 0,
              "inspect": 0,
              "submit": 0,
              "test": 0
            },
            "withheld": 0
          }
        ]
      }
    ]
  },
  "campaignId": "fc-0c7e7c67a231",
  "counts": {
    "attempts": 40,
    "cells": 34,
    "certifiedPasses": 13,
    "exams": 2,
    "graded": 20,
    "invalid": 2,
    "laneErrors": 15,
    "lanes": 17,
    "modelFails": 7,
    "withheld": 3
  },
  "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": 2,
      "countable": 2,
      "counts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "@cf/openai/gpt-oss-20b"
    },
    {
      "certifiedPasses": 2,
      "countable": 2,
      "counts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "cli/claude-sonnet"
    },
    {
      "certifiedPasses": 2,
      "countable": 2,
      "counts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "cli/codex"
    },
    {
      "certifiedPasses": 2,
      "countable": 2,
      "counts": {
        "certified_pass": 2,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "openai/gpt-oss-120b"
    },
    {
      "certifiedPasses": 1,
      "countable": 2,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 1,
        "withheld": 0
      },
      "lane": "codestral-latest"
    },
    {
      "certifiedPasses": 1,
      "countable": 2,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 1,
        "withheld": 0
      },
      "lane": "minimax/minimax-m3:free"
    },
    {
      "certifiedPasses": 1,
      "countable": 2,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 1,
        "withheld": 0
      },
      "lane": "mistral-code-latest"
    },
    {
      "certifiedPasses": 1,
      "countable": 1,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 1
      },
      "lane": "nvidia/nemotron-3-super-120b-a12b:free"
    },
    {
      "certifiedPasses": 1,
      "countable": 1,
      "counts": {
        "certified_pass": 1,
        "invalid": 0,
        "lane_error": 1,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "openai/gpt-oss-20b"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 2,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "cli/cursor-grok"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 0,
        "withheld": 2
      },
      "lane": "cohere/north-mini-code:free"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 4,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "magistral-small-latest"
    },
    {
      "certifiedPasses": 0,
      "countable": 2,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 2,
        "withheld": 0
      },
      "lane": "ministral-14b-latest"
    },
    {
      "certifiedPasses": 0,
      "countable": 2,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 0,
        "model_fail": 2,
        "withheld": 0
      },
      "lane": "ministral-8b-latest"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 4,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "mistral-medium-latest"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 4,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "mistral-small-latest"
    },
    {
      "certifiedPasses": 0,
      "countable": 0,
      "counts": {
        "certified_pass": 0,
        "invalid": 0,
        "lane_error": 2,
        "model_fail": 0,
        "withheld": 0
      },
      "lane": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free"
    }
  ],
  "perExamMatrix": {
    "exams": [
      {
        "envId": "vvdex.annotation.structured-data-1",
        "evalId": "fr-20260907-15d51112",
        "family": "annotation",
        "short": "Annotation \u00b7 structured-data-1",
        "title": "Record normalization and duplicate QA"
      },
      {
        "envId": "vvdex.annotation.text-labeling-1",
        "evalId": "fr-20260907-0626d700",
        "family": "annotation",
        "short": "Annotation \u00b7 text-labeling-1",
        "title": "Intent and entity annotation"
      }
    ],
    "rows": [
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 2,
        "lane": "@cf/openai/gpt-oss-20b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 2,
        "lane": "cli/claude-sonnet"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 2,
        "lane": "cli/codex"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 2,
        "lane": "openai/gpt-oss-120b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "codestral-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "minimax/minimax-m3:free"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "mistral-code-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/0 + 1 withheld",
            "withheld": 1
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "nvidia/nemotron-3-super-120b-a12b:free"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 1,
            "text": "1/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 1,
            "passes": 0,
            "text": "0/0 + 1 lane error",
            "withheld": 0
          }
        },
        "certifiedPasses": 1,
        "lane": "openai/gpt-oss-20b"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 1,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/0 + 1 invalid",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 1,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/0 + 1 invalid",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "cli/cursor-grok"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/0 + 1 withheld",
            "withheld": 1
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/0 + 1 withheld",
            "withheld": 1
          }
        },
        "certifiedPasses": 0,
        "lane": "cohere/north-mini-code:free"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "magistral-small-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "ministral-14b-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 1,
            "invalid": 0,
            "laneErrors": 0,
            "passes": 0,
            "text": "0/1",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "ministral-8b-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "mistral-medium-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 2,
            "passes": 0,
            "text": "0/0 + 2 lane errors",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "mistral-small-latest"
      },
      {
        "cells": {
          "vvdex.annotation.structured-data-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 1,
            "passes": 0,
            "text": "0/0 + 1 lane error",
            "withheld": 0
          },
          "vvdex.annotation.text-labeling-1": {
            "graded": 0,
            "invalid": 0,
            "laneErrors": 1,
            "passes": 0,
            "text": "0/0 + 1 lane error",
            "withheld": 0
          }
        },
        "certifiedPasses": 0,
        "lane": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free"
      }
    ]
  },
  "recordCount": 6,
  "records": [
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.structured-data-1",
      "evalId": "fr-20260907-15d51112",
      "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
      "recordDigest": "d074d0dd0c2e7a855207c95b302bfcb08bc7a092265803601f8a19c28ea6a2bb",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.structured-data-1",
      "evalId": "fr-20260907-29701401",
      "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
      "recordDigest": "4d674e21c58427397f8b9b5918e587f9111427e72eaa2281bb7d0fcfa7bd7848",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.structured-data-1",
      "evalId": "fr-20260907-7c027ac8",
      "examFingerprint": "5d6b3708dba2cdf08218a44997058413d0a3aae738299bce8c346bb2b309fc28",
      "recordDigest": "4eb1c3548b8b094ded1a52d98f0179131be2555b821d9eea7b5fc519b0f684ce",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.text-labeling-1",
      "evalId": "fr-20260907-0626d700",
      "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
      "recordDigest": "1bef7c56731085366e1774b7eaef75d81f12d3b2afeb9276f028c23ae441b421",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.text-labeling-1",
      "evalId": "fr-20260907-5b08f93e",
      "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
      "recordDigest": "b8a5d79aaefa53e7c5be413210df59f8d994e411b9f71ccaaea2b78a135503e0",
      "retired_for_future_evaluation": null
    },
    {
      "canonicalBytesPublished": false,
      "certificationReceiptSha256": null,
      "disclosureClass": "answer_key",
      "environmentId": "vvdex.annotation.text-labeling-1",
      "evalId": "fr-20260907-f7b0a094",
      "examFingerprint": "e1abe51398b2d950a3415678aaf8ec796420dfb05ea89b3b64afd33c1e345fd3",
      "recordDigest": "cb855a9a633a7a6172ac2dbab698cf69f3de94376eca921aad3451254036599e",
      "retired_for_future_evaluation": null
    }
  ],
  "refused": 0,
  "retired_for_future_evaluation": {},
  "schema": "vvdex.forge.campaign-summary/v1",
  "suspectRecords": [
    "vvdex.annotation.structured-data-1",
    "vvdex.annotation.text-labeling-1"
  ],
  "withheldByHarness": 0
}