Comment by acjohnson55

There's a better piece on this, which I read several years ago. The core of the problem isn't the use of the mean, which is what most of the comments are picking up on here. The problem is the Air Force was designing for the mean of more than a dozen measurements, which weren't highly correlated. In that case, even if your distributions are Gaussian, you're going to have vanishingly few samples that are on the peak on all of them, which is what the non-adjustable cockpit design optimized for.

This is related to the "curse of dimensionality" https://en.m.wikipedia.org/wiki/Curse_of_dimensionality