Multilayer Combinatorial Fusion for Contextual Value Alignment

The MCF-CVA framework uses an expansion-and-reduction (EAR) process to align models with diverse ethical pluralism. By fine-tuning multiple moral agents and using combinatorial score and rank fusion, the system captures complex moral contexts better than single-reward RLHF approaches.