I am not sure the assumption is correct. Look at my example...For that be true the training DMI increase must be exactly the same amount as a potential game shape DMI loss which feel highly unlikely.Until I hear anything else I will consider the assumption incorrect..