The code includes de Campos-style pruning for fNML; however, the logic behind the pruning is completely undocumented. Also, if it is actually effective, it could be published as a part of a larger study.