During unittests, I accidentally misspell the dataset names all the time. The current behavior is that an empty data structure is returned, which works fine.
I would argue that it is way better if we raise an error then, and also reflect this in an expressive error message. This should also be specified in the tool specs. @AlexDo1