[fperez_org]

Indeed, and communicating that clearly and effectively is something that we don't always succeed at. This is the reason why, on many occasions I had to resist fairly vociferous requests that we split up the parallel computing code from the shell, and also all the clients.

In the future that may make sense, once our internal protocols and APIs are so stable that we can do so without breaking functionality just as a way to help decouple development. But as a way to think about computational workflow, these pieces are all part of a coherent vision. It's just one that we don't always communicate clearly :)