Occam's razor is: algorithm tests are specialised IQ tests, and that version of IQ is actually a very reliable indicator of performance (compared to the other signals available).

I'm not exactly sure I buy it, but it seems like the simplest explanation.