> I've seen it happen that a move wasn't considered as a top move by the engine, but once played the engine realizes it's actually crushing. Something about the heuristics used to prune the vast search space can make it miss sacrifices or seemingly sub-optimal moves that temporarily weakens the perceived position but has a huge payoff in the end. But humans find them.

Just to observe, humans display exactly the same phenomenon of ignoring a move before it's made while still being able to realize, after it's made, that it was very strong and ignoring it was a mistake.