When you want to scale and be predictable, centralized execution always wins when the degrees of freedom are known. 1/
Just like if a broad product line wants a consistent interface or data collection, then centralizing those work best.
In practice, a usual outcome is just randomness because few things really have highly variable solutions. More decision makers do not converge naturally to a good solution.
yes it is also possible with stellar execution that anything can work.
This thread is about trying to be typically optimal w/o relying on outliers. // END