Fair point! What I mean to say is that there’s a lengthy calculation behind each sub-problem that determines its value, and that lengthy calculation needs only to be performed once. Any other times you encounter a sub-problem that you’ve seen before, its value can be found using the memoization array.