[ramblenode]

The problem with your reasoning is that you expect the interview to mirror job requirements rather than select for job performance. In many cases the best instrument to measure the latter will resemble the former, but there's nothing intrinsic about the relationship. If giving someone a brain teaser or having them recite trivia provides a strong signal for job performance, then it makes sense to use these instruments. You could argue that they don't provide a meaningful signal (which I think Google may have discovered with the brain teasers), but that is a separate discussion.

Something else to consider is that the interview is optimized to select for true positives and reject false positives at different rates. It's been discussed elsewhere that for a company like Google avoiding false positives is much more important than finding good candidates. So it may be the case that some instruments like trivia recitation provide the right signal at the intersection of the optimization curves. The fact that many (even most) qualified candidates score poorly on these instruments doesn't impugn their utility; their primary goal isn't to identify good candidates but to filter out bad ones.