At the moment, the model formula provided to experiments is processed with patsy. This is no longer under active development.
We should switch to using formulae as a future proofing step. It also allows for hierarchical / random effect type models, should we see fit to use those in the future.