{
  "benchmark": "robot-supervisor-cross-benchmark-100-v0.3.1",
  "results_recorded_on": "2026-09-04",
  "measurement_run_date_independently_verified": false,
  "task": "Terminal success/failure classification of object-manipulation task/video pairs",
  "accuracy": {"correct": 89, "total": 100},
  "composition": {"FAILURE": {"success": 16, "failure": 16}, "RoboFAC": {"success": 17, "failure": 17}, "ViFailback": {"success": 17, "failure": 17}},
  "sampling": "Fixed-seed stratified sampling by dataset, task family, and label",
  "seed": 20260903,
  "labels_used": "Scoring after inference only",
  "task_specific_fine_tuning": false,
  "not_validated_by_this_metric": ["evidence quality", "diagnosis", "progress", "probability calibration", "recovery guidance", "safety", "live robot control"],
  "public_demos": "Two selected examples from this benchmark; not a representative random sample"
}