promptdojo_

Train/inference skew — step 3 of 7

Two 'identical' normalizers — one derives stats from data, one hardcodes remembered round numbers. What does this print for the same user?