- Now seems to be per protocol, rather than per well, which seems unlikely to be the fastest way - May need to use `threadpoolctrl` to stop numpy parallelisation or similar https://github.com/pints-team/pints/issues/1356