A predicate method (or a query method, for that matter) defends encapsulation. The existence of the method says nothing about how the target object derives an answer. It could have the answer in a hash, it could calculate it lazily, it could delegate to another object. Whatever. The point is that the client doesn't know. Coupling is kept down. And, assuming that the question lies within the province of the target object to answer, cohesion is kept up.

I see that what you're really arguing for is reducing the need for "can you do this? yes? then do it" types of protocols. Fine; I agree. But in arguing that point, you're either misunderstanding or misrepresenting encapsulation.