Unfortunately, those metrics are very focused on the 0.1%, if not the 0.01%.

Like a sorting algorithm which is O(n) on nearly-sorted input - the utility is limited.

? That’s a common use case for sorting though.