Kevin - i tried to take what you said above and turn it into a dag or rather a selection diagram. this diagram also makes clear i think, what you would need to measure in your target population to transport your findings more widely. There are a couple of tricky things about this - the unit of analysis of the mediator isn't the same as the outcome (at least the way i've drawn the dag) - which probably complicates estimation but the concept, i think, stands. Also just want to be clear that the transport issue for us has to do with what kinds of data we will need from the target population to formally infer in the target population when the study was not done in the target population...