Models only perform on-server computation if the output artifacts are missing.

[tshead2]

Currently, we assume that when we finish a model it must compute its output artifacts from its input artifacts. Instead, we should check to see if the output artifacts already exist (i.e. were uploaded from a remote analysis), in which case there's nothing more to do.

[tshead2]

An alternative would be to allow external clients to set the model state from running to finished themselves. In fact, if we're doing external computation, we're going to need to set progress and messages, so this would make external computation work just like internal computation.

Currently, clients start internal computation by changing the state from waiting to running, so if we go this route, we'll need some other way of doing it, i.e. going back to a dedicated API for starting internal computation.