Reproduce “Towards Robust Long-Lived Agents: Benchmarking Experience Lifecycles in Evolving Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shijue Huang, Hangyu Guo, Hengheng Zhang, Guo An, Zhitao He, Qi Tian, Yiren Feng
ReScopeBench evaluates how agents maintain and use saved experience across changing interactive environments. It distinguishes structural facts from reusable rules and scores task outcomes separately from saved content checked against private current-version references. Across 352 tasks in 18 environments, Gemini-4-Argon scores 81.12 with experience versus 25.86 without; selected-cohort replacement coverage reaches 64.63% by task four, while still-valid coverage stays above 88%, a schedule-dependent result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.