KPIs

Sensitive-intent false-positive rate

Share of SensitiveIntentDetected tags that a human reviewer subsequently confirmed were not actually risky.

Formula

Reviewer-overturned flags ÷ total SensitiveIntentDetected flags × 100

Illustrative target

Directional only; too low a target risks tuning detection down rather than tuning precision up — read against KPI-G007 before optimising