I find that the question of a Consumer creating a persistent vs. a transient entity to be a little troubling. Isn't the Consumer primarily interested in getting a handle to an _entity_ (be that transient or persistent), and then using that handle to direct subsequent requests to? Could we not consider persistence and transience to be entirely Producer concerns? I understand from the sequence in the spec (pp. 6-8) that the only supporting scenario for the Consumer calling "createTransientEntity" is out of an effort to maximize the efficiency of the conversation with the Producer. This is a worthy goal, of course, but the Producer should advertise its interaction behavior in the meta-data, right? Presumably the Producer uses the meta-data to hint to the Consumer the way(s) the Consumer SHOULD structure invocation requests efficiently. I'd love to see some more convincing scenarios to support pushing the persistent/transient aspect of the conversation out to the Consumer. Otherwise, I'd love even more to limit our specification to a single "createEntity" call. Alan