#expected

The Expensive Model That Did Not Beat Size

We split DeFi failure risk into four mechanism families and fit hazard and severity models under a pre-registered standard, trying to beat a one-variable champion, protocol size. Mostly we lost, and the loss is the result. What passed reshapes expected loss: when protocols fail, the loss grows far more slowly than the protocol. This is the lab report, with the same-day adversarial corrections attached.