Prépublication · 2026
AI-Driven Reproducibility and Analytical Robustness in Scientific Discovery
Shicheng Liu, Krista Opsahl-Ong et al. — Monde
Source : arXiv
Résumé
Reproducible and analytically robust research findings are essential to scientific progress, but evaluating these properties typically requires costly manual effort. We investigate whether LLM agents can help automate reproduction and robustness analysis, using a medical case study grounded in the Surveillance, Epidemiology, and End Results (SEER) database, a widely used U.S. cancer registry underlying tens of thousands of published studies. First, we introduce SEER-Repro, a benchmark of 20 published SEER-based papers together with the data and resources required to reproduce their findings. S…