It seems these days opaque machine learning models means transparency.

I would prefer they would say that they're down grading their preference for user engagement, and time on platform, for what the trained model says it's original reporting.