Wow, I am entirely wrong here, huh? I thought that "demonstrably a reasonable measure of job performance" was held to the standard that it had to be proven fair (and consequently documented to have proven job impact).

EDIT: Hang on, I decided to just test the idea with the usual suspects and I don't see how you could design this in a way that works. You'd need some kind of non-contaminated testable outcome - lines of code etc. don't work - that is considered job performance which is scientifically validated in order to be able to survive the litigation that would follow the test.

I suppose the part that was wrong was that I claimed you can't run an IQ test. That's not true. You can. It just legally exposes you to a paperwork burden involving proof that it is required for job functionality - proof which you'll almost certainly fail because software engineer productivity isn't mechanically solved yet.

Is that the distinction? Listen, I've read your stuff and generally trust you and in any case I'm open to having my mind changed on this subject, but it seems like the actual thing is "Possible, but if you do it, you're going to find yourself in a massive paperwork and legal headache".