Firewalls From General Covariance

I define "horizon normalcy" as the approximate validity of Semiclassical Gravity and Effective Field Theory (SGEFT), for the description of observers that approach or cross a black hole horizon. If black holes return information, then horizon normalcy must fail substantially, at least in some global states. Proposals such as ER=EPR and nonisometric maps assert that horizon normalcy persists, so long as the Hawking radiation remains in a computationally simple state. Here I argue that state-dependent horizon normalcy - independently of the underlying mechanism, and independently of the class of radiation states asserted to guarantee normalcy - requires a breakdown of general covariance far from the black hole, or else horizon normalcy will depend on the infinite future of the exterior. This is because the radiation can be in different states at different events, all spacelike to the horizon crossing event whose normalcy is at stake. I discuss a related effect in AdS/CFT, and I argue that its resolution by timefolds is of no help here.