eval/score.json

{
  "label": "robo",
  "labels_quality": "gold",
  "dry_run": false,
  "WARNING": null,
  "rows": 150,
  "skill_accuracy_raw": 0.0067,
  "skill_accuracy_balanced": 0.1,
  "majority_class": "transfer",
  "majority_class_baseline": 0.3,
  "random_baseline": 0.0667,
  "num_candidates": 15,
  "call_accuracy": 0.0,
  "parsed_predictions": 150,
  "gt_in_vocab": 150,
  "n_gt_classes": 10,
  "per_skill": {
    "transfer": {
      "count": 45,
      "correct": 0,
      "recall": 0.0
    },
    "pick": {
      "count": 44,
      "correct": 0,
      "recall": 0.0
    },
    "place": {
      "count": 34,
      "correct": 0,
      "recall": 0.0
    },
    "press": {
      "count": 13,
      "correct": 0,
      "recall": 0.0
    },
    "push": {
      "count": 5,
      "correct": 0,
      "recall": 0.0
    },
    "twist": {
      "count": 3,
      "correct": 0,
      "recall": 0.0
    },
    "strike": {
      "count": 2,
      "correct": 0,
      "recall": 0.0
    },
    "pull": {
      "count": 2,
      "correct": 0,
      "recall": 0.0
    },
    "slide": {
      "count": 1,
      "correct": 1,
      "recall": 1.0
    },
    "fold": {
      "count": 1,
      "correct": 0,
      "recall": 0.0
    }
  },
  "note": "Pre-segmented single-skill clips -> measures naming-given-segmentation, NOT autonomous segmentation. Prompt contains no step text / gt / call."
}