Data strategy to test workflow

We can either:

• generate the data from scratch before the test
• keep a stripped down version of real data in a HDF5 file in the repo

Other ideas ?