The difference I was aiming at is that you (a library author, whether public or internal) shouldn't implement adder, you should implement add. And let your caller implement adder using add, if they require it. If you only expose adder, you limit possible uses for your library.