outputs/evaluation_example_v10.json

{
  "schema_version": 1,
  "generated_at": "2026-08-11T05:50:38.509439+00:00",
  "status": "LOCAL_FINITE_CANDIDATE_EVALUATION",
  "format": "generic",
  "dataset": {
    "name": "synthetic-documentation-example",
    "revision": "v1",
    "license": "repository-example-only"
  },
  "artifacts": {
    "items": {
      "path": "examples\\evaluation_items_v10.json",
      "sha256": "ad92f109c2f0b908652a627114ed064716887a4273ced550fad59ea4c09e0468",
      "size_bytes": 597
    },
    "predictions": {
      "path": "examples\\evaluation_predictions_v10.json",
      "sha256": "ce01d409d38ab8d0c01093dd384fd398186831df5ae83fa76dfc345688782c74",
      "size_bytes": 311
    }
  },
  "evaluation": {
    "rows": 4,
    "candidate_limit": 8,
    "metrics": {
      "effective_acc_at_1": 1.0,
      "candidate_recall_at_k": 0.5,
      "baseline_mean_min_cer": 0.25,
      "mean_min_cer": 0.0,
      "abstention_rate": 0.5,
      "accepted_rows": 2,
      "accepted_accuracy": 1.0,
      "clean_rows": 2,
      "overcorrection_rate": 0.0,
      "typo_rows": 2,
      "typo_accuracy": 1.0,
      "improved_rows": 2,
      "worsened_rows": 0,
      "unchanged_rows": 2
    }
  },
  "claim_boundaries": [
    "An empty or missing prediction abstains and preserves the input text.",
    "Overcorrection is reported only for rows explicitly labelled clean.",
    "Candidates are scored as supplied; this command performs no model inference.",
    "AJIMEE provenance is pinned only when the item artifact SHA-256 matches."
  ]
}