An issue with this is that it's inherently one-sided: you get to see the performance of the people you hire, but you have no idea what happens with the people y

An issue with this is that it's inherently one-sided: you get to see the performance of the people you hire, but you have no idea what happens with the people you reject. Because of that, you inevitably get "weird" results like seeing that your best employees often did relatively poorly on interviews (Google did a study that produced exactly that result—their best employees often were those that failed 1 or 2 of their interviews). But this is just selection bias: these are candidates that got hired! Which just means that they were so spectacular in some other respect that they were able to get an offer /despite/ poor interview performance. It tells you nothing actionable about your interview process. Really you need to take an experimental approach where you randomly give offers to people who failed the interview, and see how they turn out.