With approach 3 (unidirectional aging) wouldn't the rapid eviction problem compound in a network where most nodes are behind Tor? If addresses age out in 2 hops and Tor connections are already slower to propagate, you could end up with nodes that can barely discover each other. Seems like approach 2 might hold up better in practice since it at least preserves real timestamps within the same network type. Have you tested what the address table looks like after a few hours with each approach?
With approach 3 (unidirectional aging) wouldn't the rapid eviction problem compound in a network where most nodes are behind Tor? If addresses age out in 2 hops and Tor connections are already slower to propagate, you could end up with nodes that can barely discover each other. Seems like approach 2 might hold up better in practice since it at least preserves real timestamps within the same network type. Have you tested what the address table looks like after a few hours with each approach?