Reproduce “Not All Evidence Is Admissible: Evidence Governance for Tool-Augmented Medical Agents” 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
Runxun Zhang, Onofrio Antonio Catalano, Jinqiao Wang, Jingwei Wei
CIEG filters medical-tool outputs before they reach a frozen reasoner, using a separate Checker to compare evidence cards with the original input and admit selected cards to the Solver. On 400 REFUGE2 glaucoma cases, all three cards flipped 178 originally correct answers. CIEG achieved 0.730 balanced accuracy versus 0.637 for budget-matched random admission, supporting benchmark evidence selection, not clinical benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.