Greg was working under the assumption that every validator knows what the set of all validators is. We had been working under the assumption that each small validator set knew the validators in its set, but no others. With Greg's assumption, he can decouple contracts from the validator topology.
In Greg's model, there is a map from validators to the namespace they validate. Contract parameters have a channel sorting that includes a namespace, like "x is a channel in the B namespace on which we can receive names from the A namespace". Given that information, one can check at contract deployment time whether there are resources deployed to process the contract. If not, then there are various choices how to force/motivate validators to cover that namespace.
Greg's model is orthogonal to our model. His organizes the interior of a validator set; ours groups validator sets together.