@dinosaur (happy to ramble along)

to me, Scuttlebutt has always been intentionally focused on small-world network topologies.

Trust is highest in highly clustered areas (i.e. within a single social circle), yet efficient and resilient replication wants to bridge between clusters, that is it wants to follow the paths of least trust. There's an inherent tension there, and no optimal solutions. This is among the sacrifices I mentioned about focusing on maximizing trust during replication.

I lean towards prioritizing replication resilience and efficiency, and so does Dominic, I think. Which makes me so wary about relying on the trust guarantees of the network.

If I were to break compatibility and redesign ssb completely from scatch, I'd use a peer sampling service on a fully random overlay network, and then use a quality metric (primarily) based on feed subscription overlap to prioritize replication connections (and also do some clever scheme of gossiping network addresses based on the quality metric). That is pretty much throwing all the trust overboard, instead optimizing for replication quality. Actual ssb is much more moderate, and that's probably a good thing.