This is interesting.
As I was reading some of the responses above, I was wondering if there was any real way to determine "good" from "bad" stake, and how that may implicate witness voting.
Based on your explorations of a potential 'reputation system that actually works', and the original comment from @argsolver above -- would using a new reputation index as a criteria for running a witness be an effective way to do it? Or would another path, such as maybe applying some kind of bonus/penalty to the effective stake being used to vote witnesses work?
Regardless.. Definitely eager to hear more background / context on DPoS, as well as your (and @theoretical's ) thoughts on future improvements / different protocols.
Here I define "REDPOS" as a DPOS protocol evolution thanks to the reputation factor.