I have implemented my suggestion 1 and 2 in this branch.
We will let this branch run for a while on our node and check how addrman quality evolves. Suggestion 3 is not implemented yet, it will only be effective if many nodes implement it, so cannot really be tested on a local node - and I definitely would want to do some simulations for that before actually PRing it.
I think that a simulation wouldn’t achieve much realism because the current state of the network is not very natural:
Any possible increase in incoming connections is going to be dwarfed by LinkinLion or similar spammers, which make up to 10 connections each second to reachable nodes, resulting in the NodeId counter increasing by almost a Million each day on my nodes.
Given that these “attacks” have not had much of a negative impact on the network (or were even noticed by most), I don’t worry too much about one more feeler originating from a node every 30s.
Interesting -I will report back if I see similar.