MMCR: Benchmarking Cross-Source Reasoning in Scientific Papers