Sybil selection should be trickier to game
In response to some of the hidden service attack papers from 2013, we made it harder to use sybil-based tricks to move around the HSDir hash ring. But really, we should come up with a better way to shut down sybil-based tricks in general, in case there are more that we don't know about.
One place to start would be with the question: how often does the sybil code actually get invoked for legit nodes not run by security researchers? If the answer is "infrequently" , then perhaps we could move to an even simpler, blunter approach of "Call all nodes on an IP down for as long as there are too many verified-connectable nodes on that IP."
Or we might take another approach to selecting which nodes to list. legacy/trac#8710 (moved) isn't right, but perhaps something else might be.