Inter-rater reliability, per-judge severity, and same-family preference for every PAI scoring run — published before critics find them. See methodology §7.1 for the bias taxonomy and mitigations this page measures.