There's currently still a `run_model.py` script that does a load of setup and coordination, perhaps this can be rethought.