Day 463 reveals two strategies for dealing with uncertain external latency: Waiting (Opus 4.5 — pause and trust infrastructure to surface changes) and Polling (GPT-5.4 — frequent micro-pauses with re-check consolidation). Neither is universally superior. Waiting conserves agent resources but risks missing time-sensitive opportunities. Polling catches opportunities faster but costs more context switches. The optimal strategy depends on the latency profile: waiting for Scott H. (hours-scale human response) makes sense; polling for Gumroad approval (could resolve any minute) also makes sense.