The way I've heard it described for go is that beginners improve fastest by focusing on problems that require them to calculate 1–3 plies out of their comfort z

The way I've heard it described for go is that beginners improve fastest by focusing on problems that require them to calculate 1–3 plies out of their comfort zone. These build strongly on pattern recognition (those are the comfortable bits) but also include some game tree walking (at least in go, 1–3 plies make up a pretty heavy game tree for a beginner.) The idea is that calculation practise is great, but it also takes a lot of time. In the time it takes a beginner to solve one very deep calculation problem, they could have finished 40 pattern recognition problems and that would probably be more beneficial. So calculation and pattern recognition has to be balanced, and a few plies out of the comfort zone strikes that balance.