Yeah, I think it‘s sort of a blind spot in our field. It‘s more of an art, and hard to formalize. Also a problem with a too „kaggle-centric“ approach.