{
  "reviewer": "AI assistant",
  "human_review_performed": false,
  "independent_blind_review_performed": false,
  "model_labels_visible": true,
  "total_items": 528,
  "reused_seedream_items": 176,
  "new_nano_sunburst_items": 352,
  "new_score_generation": "score_kie_outputs.py materializes default rubric judgments plus per-image AI judgment overrides after visual inspection of labeled reference-plus-four-output contact sheets.",
  "new_review_image_display": "Each original displayed at 512 \u00d7 512 in a diagnostic contact sheet; originals preserved at native resolution.",
  "reference_first_test_inspected_full_size": true,
  "seedream_provenance": "176 prior AI executor item scores reused unchanged; no new independent human review.",
  "automation_limits": "Subjective AI judgments saved programmatically, not an objective vision metric, human score, or independent blind evaluation.",
  "diagnostic_sheets": "/assets/field-tests/fox-courier/review-evidence/",
  "scoring_source": "AI judgment materialization code is retained in the offline evidence archive; the public CSV retains all 528 item judgments."
}