I was thinking about it a fair bit lately. We have all sorts of benchmarks that describe a lot of factors in detail, but all those are very abstract and yet, those do not seem to map clearly to well observed behaviors. I think we need to think of a different way to list those.