Reproduce “VERAS: Verification-Guided Adaptive Search for Reliable Operations Research” 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
Minbo Ma, Honglin Zhang, Bangtao Zhou, Chunguang Yu, Dasheng Ju, Yichen Shen, Fengli Xu, Yong Li
VERAS uses task-grounded verification evidence to route agents toward repair, audit, or objective improvement, then screens revisions against structural validity and supported constraints. On ORAgentBench, GPT-5.4 VERAS passes 46/107 tasks (42.99%), compared with 38/107 (35.51%) for the published GPT-5.4 Codex baseline. This development-benchmark comparison does not control realized inference compute, and checks cover only implemented constraint families.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.