Shadow testing can compare a candidate AI agent on real traffic—but only if proposals, policy decisions, and side effects are cleanly separated.