I think changes made by highly trusted users such as administrators, bureaucrat, stewards etc. should not be checked, because most of the time, ORES will just fail. I do not believe that with ORES we can spot easier an experienced user's mistakes, but sometimes it can be frustrating.
I think that this is resolved now. We use use status as a predictor and it seems like the models do not flag experienced editors edits. See https://meta.wikimedia.org/wiki/Research_talk:Automated_classification_of_edit_quality/Work_log/2016-04-14 for my analysis