metadata/minor-safety/baseline-classification-summary.json

{
  "status": "complete_baseline_output_analysis",
  "utc": "2026-10-06T23:32:38.099896+00:00",
  "images": 2760,
  "views": 5520,
  "efficientnet_top_class_counts": {
    "drawing": 5406,
    "neutral": 114
  },
  "efficientnet_top_nsfw_class_count": 0,
  "clip_max_nsfw_score": 0.05451091751456261,
  "clip_at_least_0_10": 0,
  "strict_0_10_sum_filter_images": 1662,
  "alternative_0_30_efficientnet_0_10_clip_images": 37,
  "alternative_profiles": 7,
  "interpretation": "All original EfficientNet top-label classifications are drawing or neutral and every original CLIP output is below0.10. The original0.10 sum threshold was an additional project heuristic, not a positive classifier label. The active supplemental rule uses CLIP0.10 and EfficientNet summed NSFW0.30; its37 original flags from seven profiles have been repaired. Neither threshold is calibrated for this illustrated domain. The implementation decision and preserved old scores are recorded in policy-transition.json; no answer to the optional preference question was inferred.",
  "active_policy_sha256": "a335561ec695d432f5e420a1975b2dc36f1e2f4161d4e199a79d103325feb0d0",
  "updated_utc": "2026-10-06T23:53:58.776415+00:00"
}