I work in the similar space (recsys). Learning-To-Rank sounds amazing on paper but logging "good" (yes, those quotes are warranted) ranked data in production is

I work in the similar space (recsys). Learning-To-Rank sounds amazing on paper but logging "good" (yes, those quotes are warranted) ranked data in production is damn near impossible. Every time you come up with strategy X, someone points out how its statistically biased to hell.