I like this idea! To build on it, the schedules could be non-random (systematic with random offset) and test combinations of substances in a way that maximises information both in the cases where substances are independent and when they have suspected interactions.