ttwos/

Anon-T2S-Bench-MR

A benchmark of 500 multi-choice questions for evaluating multi-hop reasoning over implicit/explicit structures extracted from scientific text. Each sample pairs a paper-grounded text passage and figure with a question, ground-truth answer, reference reasoning chain, and reference structural frame.

	Schema




	

Field Type Description

pid string Sample identifier

paper_title string Source paper title

topic string Research topic… See the full description on the dataset page: https://huggingface.co/datasets/ttwos/Anon-T2S-Bench-MR.