InquiTree Benchmarks Agents on the Full Scientific-Inquiry Loop via Paper-Derived Research Trees
arXiv·low signal
InquiTree (arXiv 2606.09550) formalizes scientific inquiry as interactive Research Trees — DAGs capturing dependencies among hypothesis formulation, study design, result interpretation, and belief updating — to evaluate whether agents can reason through an investigation rather than answer one-shot questions. It targets a gap most agent benchmarks miss: sustained, dependency-aware reasoning over a multi-step inquiry. Relevant to anyone evaluating 'deep research' agents beyond single-query retrieval.