After reading the OP I immediately thought of your standard hiring advice and wondered if the folks asking math questions aren't on the right track after all. (Thank you for your persistence posting it, btw.) At least half of my daily work requires learning new things on-the-fly, and I bet in 5 years most of my technical knowledge is turned over anyway--except the math and CS fundamentals. In programming aptitude matters way more than current knowledge. I think IBM discovered the same thing in the 60s when they were hiring chess players and crossworders. So if math puzzles serve as a legal(ish) proxy for an IQ test, aren't they just what interviewers should ask? Of course I'd rather have someone who knows Ajax and 3NF if I can get it, and not lose any time to up-front training, but that may not be realistic, and those things can be learned if the person is smart. Particularly for an entry-level hire, I'd rather hire someone who can solve graph theory problems than knows the command-line options for git.
Math questions are probably an okay substitute for "general IQ" testing for the majority of population, moreso than most other question categories (math is literally just logical reasoning). But, it's not a truly general skill, it's a specialized discipline--and that means the practice of using it as a screen is going to fail hard when you try to apply it to someone with specialized training. Which is what happened in the article.
