Likelihood penalty seems to be reasonable, and adding constraints seems to favour parameter values closer to the constraint mean. More tests should follow...Tests with toys show spreads and correlations between parameters that are consistent with the constraints applied