Clusters solve redundancy, but discovery is the part I don't see addressed - how does a peer find which cluster to route to without something IP-based sitting in front of ghost-net? Feels like these designs move the bottleneck rather than remove it.