[thaumaturgy]

Yeah. I think some of that would resolve itself as the network grew. Otherwise, whenever I've spent much time thinking about it, I've figured that it should be possible to have a nearest-neighbor node notification protocol.

Node A declares that it's quitting to Nodes B and C (or Nodes B and C are suddenly no longer able to poll Node A); Nodes B and C both check the chunks that they know Node A had, and see if there are enough redundant copies of those chunks in the network.

There's a lot of "and then magic happens!" in this idea, but I'm pretty sure it's do-able. The worst part of it by far is doing node searches on a completely decentralized structure.