Hi @wakandan, the way rasa test nlu tests is that if you do not provide a model, it will use cross validation; if you do provide a model, it will evaluate that model on the provided data. So what I would suggest in this case is creating subfolders (or just files in your data folder, but adjust the paths if necessary:
rasa train nlu --nlu data/train
rasa test nlu -m models/<timestamp> --nlu data/test
(you can also just use -m models to get the most recent model in the directory instead of putting in the timestamp)