For stability predictors fx. FoldX we apply a sign flip, s.t. we can conduct black-box maximization and maximize stability. In the (current) Rosetta black-box implementation we do not apply a sign flip.
Required addition to the black-box inner function.