I suspect (and have seen some evidence) that the interviews he aced were algo-based. Doing well in these is very repeatable, with low additional effort. Behavioural are much harder to do at scale.