Game analysis using statistics

Yes, the idea is there, and so is the solution: Let's select the games to be analyzed that did not end in a time loss and where the player has a strong rating (1600 ) today? We can also limit the depth of the moves to be analyzed (the first 30 moves) and call the result "opening learning".

I would add one more thing to what Velvel mentioned - lightning, where it is relatively difficult to find quality.