Very supportive of correlation penalties! I’ve previously advocated for starting the quadratic inactivity leak at a much higher percentage (e.g. 90), but I think a change like that comes under;
so I think a design around relative misses is more sensible. However I think we could explore a non-linear proportional penalty where you suggest:
I think @vshvsh 's comment is pretty astute, and suggests a decent amount of retrospective analysis and prospective modelling would help with setting effective parameters for such a penalty.
I also want to highlight that this could hopefully be an effective way to put the finger on the scales for client diversity. We have had numerous liveness failures from clients, (and thankfully no safety failures) and if we tweaked the correlation penalties such that a double-digit% liveness failure resulted in a month or more of lost rewards, that would have an impact on operator behaviour in my opinion.
One ‘parameter’ that is probably closely intertwined with this one is the ‘socialised penalty’ that the whole network faces when participation is <100%. I think this feature is important and worth protecting, to keep operators in a win/win mindset with respect to uptime rather than win/lose, but some adjustment such that those who are offline during the outage get hit harder than those who remain online through the outage is probably not a bad thing.