> This might be true, but it's really really hard for someone who doesn't work with them day-to-day to distinguish.

It's really not (in mathematics, the field under discussion). It's not hard to look at the CV of a graduating PhD student in math and tell whether they're a semi-reasonable candidate for the Clay Fellowship, or the other fellowships listed above.

Among the small group who makes that cut, of course you need to rely on letters of recommendation, expert assessments, and so on. But the initial cut is fairly straightforward.