I think it's complicated by the fact that the better the company, they better they are at mentoring even "bad" hires. I.e., performance review isn't normalized across all companies. Google might think their hiring process is great because a lot of their hires do well in their performance review - but maybe that has more to do with the engineering culture and mentorship in the company, and less to do with their ability to pick future winners.

Perhaps though if one looks at the performance review weighted to a review date as soon after the hiring, and sort of tail off how much weight is given over the coming years it'd make sense.

I've just seen a lot of bad hires get better with time. And some good hires lose steam or become jaded and perform worse over time.

Thus, as messed up as it is, one of the keenest signals is tenacity in the end. And ironically tenacity in preparation for interviews (that Leetcode grind) is somewhat correlated with performance reviews in the long run and perhaps correlated with utility to the company.

That said, "decent" engineers in the long run that don't think outside the box but check their performance review boxes slowly bleed a company's ability to disrupt. I.e., performance reviews need to be expanded a bit to think about the health of the overall company and not just the employee performing their function.

As a large tech organization, I'd prefer a complex distribution of performances over an even distribution any day - because value to tech companies is still influenced by more black swan event type of innovation. But then I have a high "beta" (variation) in terms of what I'm looking for. Companies with more of a utility emphasis perhaps are better served by a hiring process that is "reliable" with a lot of false negatives.