the real problem is not that we can't measure productivity, but that we can't even fully define an abstract productivity metric—what does "productivity" even mean?
we can come up with some notion that talks about the net effect of an individual (the "wins about replacement" of programmers), but that tells us absolutely nothing about how any given individual achieves that; hell, the net effect of any given individual is presumably a function of the entire context and the rest of the org, not just of the individual!
alternatively we can try to define more direct notions of "productivity" even if we can't measure them, but those notions end up varied, multidimensional and, again, painfully context-specific—it's absolutely a useful thing to think about, but it does not let us pin down what a "top 1%" engineer "really" is, or even if that's a meaningful notion
Individual productivity exists.
Maybe it's easier to measure groups' productivity? Probably.
"Business impact"? I don't think so, that later concept seems much more arbitrary. But feel free to look for the keys under the lamplight. If you choose that metrics, you're not going to retain many extra productive people anyway.
The old problem: judging the work of an expert is very difficult if you lack comparable expertise. I can give you advice, but I can't make you smart to accept it. How could you tell if I'm a genius or an overconfident asshole?