Reproduce “Safe, Persistent, and Evolving Agent Harness for Understanding Partially Observable Worlds” 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
Yisen Gao, Yue Guo, Qing Zong, Yiwen Guo, Yangqiu Song
E-Ledger combines state tracking and a code gate with WorldAbduct, which tests hypotheses about hidden tool effects before adding supported rules to its ledger. It diagnoses failures and probes explanations. In a single run on World of Workflows’ 20-task test split, GLM-5.3-Flash reaches 75% safe task completion versus MemoHarness’s 65%. This result is limited to that benchmark and setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.