selecting survivors of comparative study without understanding where the performances come from gives you no guarantee of future performances - which is more or less the issue with all these complex multivariate regressions that journalist, startups and venture capitalists insist calling AIs.