Notes
What we measured, and what changed because of it.
What a second review catches when it asks different questions
AI-drafted variant definitions that had passed one adversarial review were re-audited on different axes. 8.7% of the sample was defective.