Some of you may remember a brief discussion at IETF 108 where I discussed a difficult privacy tradeoff in QUIC-LB. I didn't get much solid feedback at the time, though a few people said they would think about it.
In the absence of guidance, I've thought it through and proposed a solution here: https://github.com/quicwg/load-balancers/issues/12#issuecomment-718222873 and that solution is to do nothing. Anyway, I believe the comment captures the sacrifice that choice makes. Comments are still welcome. Martin
