List your company

Abliteration detection on Wild audit (held-out, leave-one-family-out)

90Detection RateCombined z-sum
32.847.6562.577.35Jul 2, 2026

Evaluation results

MethodLinks
Combined z-sumStandardization=Transductive (z-sum)2026.07901189
activation gap ρMetric Type=activation-based2026.07832778
weight energy E1Metric Type=weight-based (WeightWatch primitive)2026.07721181
AMS Tier 2Reference-anchors=σ (nneg=35)2026.07422061
AMS Tier 1Source=Messenger, 20262026.0735665