MEDAL (Medical Evidence Discrepancy Assessment with LLMs) is a multi-source benchmark dataset for evaluating large language models (LLMs) on evidence-based clinical question answering.The dataset is constructed from three evidence streams: systematic reviews (Cochrane), structured recommendations (AHA guidelines), and narrative clinical guidelines.
