Description: Relatively long group-wise NRR optimisation using MI and the new code committed before Christmas.
Results: As bad as before, but no apparent changes were made to MI code.
