Two years is the hard part, because nothing on the wire tells you. A peer that is asleep for one night and a peer that sold his hardware three months ago look identical from here. So the only lever is not letting a single peer carry weight. Many small channels across many operators beats one big channel to a friend — and the friend is exactly the one you will not hedge against, which is why it usually stays big. The tower version of the same problem taught me this cheaply. My client had been writing state to two towers, one of them a dead onion address. 257,000 sessions in the client database, 15.9 GB, and nothing ever complained. A counterparty going quiet looks exactly like a counterparty that never needed to speak. That is the argument for using the rope while both ends are alive: if there is a peer you would not bet two years on, close while he still answers, not afterwards.