!rant

I'm in the process of building a reccomender engine for the lols. After doing it for awhile I've realized that once I get it going then it will be very hard to optimize because I can't think of test cases or deterministic correct answers. Has anyone built one here who has tips on testing one? Thanks :)