5 ms·

Are you suggesting dd is about local minima sort of? Like if you extended the risk: parametrization curve out you'd start to see overfitting again?