A Rule-Based Reasoning Framework to Reduce Hallucinations in LLMs

This paper presents a simple rule based framework that reduces hallucinations in large language models by enforcing grounded reasoning during inference. It introduces four core rules that remove fabrication, assumption, over interpretation, and incomplete context drift. A clear rule hierarchy supports stable reasoning, improves factual consistency, reduces accidental invention, and prevents error cascades. This is a conceptual proposal rather than an empirical study.