1) No, there is a set interval of 60 seconds (which is also the transaction failure duration) where eligible nodes are valid. So say that the time now is 00:01:00, a set of nodes are valid, at 00:02:00 a different set of nodes are valid based on the endorsements made in the interval 00:00:00 -> 00:01:00. This should allow even the most latent of nodes to keep pace.
2) Hmm I'm not sure how you have come to that conclusion. If nodes go offline, then come online again later, they will receive endorsements again, which will increase the eligible set of nodes again.
Yes I'm aware of that, the FBA algorithm is the replacement for that broken one.
1) I suppose it depend on what bounds the latency is under. I agree that 60 latency is hilariously long for nominal operations, but its always the edge cases which cause the problems
2) If nodes go off-line never to return, doesn't that reduce the set eligible set to 0 on the limit?
Regarding FBA - I thought that was the broken design, not the new one?