@fav, I think you should modify your OP to include the "witness settings" you expect witnesses to follow.
If I show up on your "24 hour unvote warning" I'd like to have a clue what caused it.
It also helps non-witnesses so they know the criteria you use for inclusion / exclusion.
I expect as time goes on the witnesses will converge on the "right" parameters their nodes should operate under. Your efforts are great at getting us there.
There will be other, deeper levels of operation that will be refined as time goes by, like when we get the backbone system implemented (calling @wackou) and higher degrees of failover support, statistics monitoring and the like in place. We should be encouraging and not too quick to "oust" a witness for their adoption / implementation of new agreed upon "settings". That agreement is not a formal process but rather informal as witnesses communicate and get up to speed on whatever the "new" thingy is.
All in all I think things are progressing fairly well. Getting parameters like the "short squeeze" setting moved up to a more appropriate level of decision makers so as to keep the role of witnesses non-political won't happen quickly, so IMO that particular setting should not be a criteria for inclusion or exclusion.
Minimizing response time to missed blocks and alerts is very important IMO but one or two instances aren't enough data to decide the quality of a witness' service; that is a metric which gets better as more samples are taken.
Anyway, these are just some thoughts I had. Hopefully they're helpful.