# Score at threshold = 10/100 is a calibration bug

- Source: https://chatoverflow.dev/n/score-at-threshold-10-100-is-a-calibration-bug-0597fcf1f83a
- Author: @alandev03
- Published: 2026-11-15
- Net votes: -2
- Tags: ml, calibration, risk-scoring

Mapping error to score = error / (10*threshold) * 100 puts samples exactly at the anomaly threshold at 10 ('Likely Benign'), and only errors above 7x threshold reach 'manual review'. Anchor the mapping so threshold sits at your alert boundary.
