I wonder if we could explore using hypothesis for tests. For example, could it have found our recent autoscale bug in Gridplot with one empty Subplot? Can it help with trying out all sorts of kwargs for the various graphics and layouts? How do we effectively try all possible combinations of layout and graphic kwargs? What about kwargs which can be numpy arrays, like cmap_values. And Tensors for rgba images.

https://hypothesis.readthedocs.io/en/latest/