Earlier quoted context omitted.
One has to wonder how comes that recommendations were pretty good, but then they spent millions of dollars on improving them, including highly-publicized competitions and a lot of work inhouse I assume, and not they are not.
I have movies I have already rated (i.e. watched) in the "watch next" list. Sure, it can happen that I want to watch them again, but it seems unlikely. It seems clear netflix is optimizing for something which is not just "the movie this user would like to watch next".
To that end, training the model in their users, it would not surprise me to have that be an emergent recommendation.