we believe that due to stronger emphasis on optimization and ML rather than, say, on the empirical details of web page

Thank you for your feedback. Reviewer 1: Regarding Web and data mining conferences, we agree that this work is relevant to them as well. Reviewer 2: To answer your question about domain-level modeling of change rates: absolutely! In the same vein, it is common to do it at the site level. This won't affect our RL algorithm's theoretical guarantees, but will certainly improve its empirical convergence rate.