Scientific Opinion is All We Need Now: A Deterministic Proxy Framework for Mediated LLM Consensus

Large Language Models (LLMs) have recently demonstrated remarkable capabili- ties in generating human-like text, fueling their potential to assist in complex tasks such as medical research, scientific literature analysis, and collaborative decision- making. However, uncontrolled multi-LLM interactions can lead to incoherent discussions and lack of consensus. We propose a deterministic proxy framework that mediates multi-LLM conversations, enforces a shared convergence metric, and logs each interaction in an auditable format. Our system, titled Peer Consensus, en- sures iterative alignment by requiring each LLM to explicitly report a self-estimated agreement percentage with peer opinions. We show that this structured approach accelerates consensus formation in domains such as cancer research, while pro- viding transparency and reproducibility through per-model SQLite databases. The full source code for our framework is available at https://github.com/nestoru/peer- consensus.