It’s possible that a human review system would be even worse. Have you seen the studies that show how insurance adjusters, working from the same inputs, with clear evaluation rules, have spreads of > 40% for the same case?