I would argue that Microsoft's original practice (without the 20/70/10) is actually pretty good. Have managers make subjective evaluations, merge them together

I would argue that Microsoft's original practice (without the 20/70/10) is actually pretty good. Have managers make subjective evaluations, merge them together at higher level meetings, and then work out compensation from there. There's a big cottage industry of trying to back everything up with data, to provide actionable feedback, etc., and these end up being giant time-wasting cover-your-ass exercises, which always end in an uncomfortable non-working system for everyone -- "I did the thing you asked but my review is the same as last year, why aren't I getting promoted". Mentorship and growth has to be more than just "here are your goals". Peer reviews can be okay, but only if you force people to make judgments -- "rank these three people against each other" rather than "give these people a rating 1-10 in each of these five areas". The subjective evaluation process doesn't work unless you trust your managers, though. And that invariable means that it doesn't scale.