And at scale you even have a "sampling" of sorts (even if the distribution is very narrow unless you've done something truly unfortunate in your FP code) via sc

And at scale you even have a "sampling" of sorts (even if the distribution is very narrow unless you've done something truly unfortunate in your FP code) via scheduling and parallelism.