Digital Pantheon: Simulating and Auditing Coalition Formation with LLM Agents
A multi-agent framework that reconciles factual grounding with ideological alignment by combining Supervised Fine-Tuning, Direct Preference Optimization, and Retrieval-Augmented Generation is presented, which yields a stable winner and ranking, and manifesto-anchored lineage reliably predicts real-world materialization whereas hallucinated content does not.