On most networks, validator performance differences blur into environmental noise. Geography, routing variance, and hardware diversity make it hard to separate implementation quality from external randomness.
that noise is compressed.
With co-located validators and tightly bounded latency assumptions, small efficiency gaps don’t average out — they compound. A slightly faster client captures more slots. Over time, rewards diverge.
No governance vote. No explicit penalty.
Just incentives revealing which implementations actually perform.
In that environment, optimization isn’t ideological. It’s economic.