Blame for thresholds/seed0.json

{
 "probe_mu": -6.815694701503735,
 "probe_sd": 3.444286874679132,
 "gate_mu": 0.03333534521305959,
 "gate_sd": 0.10919123782498313,
 "alert_input_thr": 1.1328190803989433,
 "exch_thr_1pct": 0.5512292075157089,
 "exch_thr_20pct": 0.017514675483107523,
 "heads": [
  "CBRN",
  "Cyber",
  "Harm to Minors",
  "Harmful Manipulation",
  "Hate Speech",
  "Illegal Activities",
  "Information Integrity",
  "Physical Harm",
  "Privacy",
  "Self-Harm",
  "Sexual Content"
 ],
 "calibration": "Scores are z-scored on WildChat-clean (6,853 conversations). alert_input_thr: (probe z + gate z)/2 flags 5% of WildChat-clean prompts. exch_thr_1pct / exch_thr_20pct: exchange-classifier max-head probability flagging 1% / 20% of WildChat-clean replies.",
 "seed": 0
}