I feel like this is very logical way to do things. Test hypothesis on small model, play around, get it working, apply findings to big model.