error type,count,% of test set
degenerate repetition (4-gram repeated ≥2x),4,0.26
source copied verbatim,0,0.0
empty output,0,0.0
severe under-generation (ratio < 0.5),26,1.71
severe over-generation (ratio > 2.0),8,0.52
near-miss (60 ≤ chrF < 90),342,22.44
hard failure (chrF < 20),306,20.08