Consistency-Guided Decoding with Proof-Driven Disambiguation for Three-Way Logical Question Answering

Implement a 'Proof-Driven Disambiguation' decoder that forces consistent logical outputs for three-way (T/F/U) questions. This should be a wrapper for existing models to resolve contradictions in RAG pipelines.

Suggested repo: LogiChain

"Stop your LLM from contradicting itself in logical tasks."

Estimated effort: 45h