I've been running it as :

session.run(to_outputs=['my_dataset'])

And the return value is an empty dict. The pipeline runs smoothly, and everything is defined in the catalog.