> It has technical merits -- it makes it easier to deal with a host that breaks the contract, because there are lots of nodes, so you just 'route around' the faulty host's output

How do you decide whether a host is faulty unless there’s consensus on what constitutes non-faulty? And how do you reach this agreement in a distributed manner without distributed consensus?

You can’t switch away from a bad actor in a decentralized system unless you have a decentralized way to establish who’s a bad actor.