Reproduce “EvoSCM: Scientific Belief Revision Through Causal Model Evolution and Experimentation” 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
Qing Zhao, Haowei Li, Weijian Deng, Sibei Yang, Pengxu Wei, Liang Lin
EvoSCM evaluates discovery in simulated worlds using competing structural causal models (SCMs) that commit to predictions before experiments. It selects interventions where models disagree, then turns prediction errors into revisions checked against prior evidence. On DiscoverPhysics with GPT-5.5, it scores 75.09% on mechanism explanation versus 51.64% for baseline, using 1,685 versus 2,206 simulator episodes. Results are limited to simulated tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.