Thanks for surfacing! Enjoyed reading the paper a lot, some good ideas and data. However, there are some major ways in which our network model is misaligned from the major success cases described in the paper. First, we allow direct connections between local peers (ex to handle within-NAT discovery), so we fall into the class of networks where R5N actually performs worse in small scales. Also, we want to ensure that attackers aren’t easily able to DOS the network by making requests that cause many other nodes in the network to perform work. R5N is comparing against recursive kademlia, which has the same abuse vector - but we use the iterative version to ensure an attacker is more bounded in their ability to waste network resources. There may be other relevant learnings, will continue to discuss in our upcoming Content Routing design discussions!