Here's the thing, get used to single big commits. Eventually, somebody is going to try to train on specific change sets. This'll enable models to learn specific authors mannerisms, idiosyncracies etc... Single large commits creates an info asymmetry boundary, which is about the only defense a creator has in a world of willful infringement to train algorithms to replace or devalue them in the market. It sucks... But this is the world we're growing into now.