But the general point about not persisting intermediate datasets unless there's a good reason to do so is still very valid - people tend to persist more things than they really need to from what I've seen