Why the average is the least useful number

One number, two populations

An endpoint that serves a cached read in two milliseconds and a payment in four hundred does not have a latency. It has two, and averaging them produces a number that describes neither.

This is not a subtlety. Blend a fast majority with a slow minority and the mean sits in a gap where almost no request actually lands — comfortably inside your SLO while a fifth of people are having a terrible time.

A mean is a summary of a population. It is only meaningful if there is one population.