Revealing the data behind the recommendation. This is what you meant, right? Any comments about how it's implemented?