really glad to see _this_ kind of analysis of these models. These type of analysis is much more practically useful than some context-less (especially with regard to time) number of "how good it is".