[2603.02798] Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification

Computer Science > Artificial Intelligence arXiv:2603.02798 (cs) [Submitted on 3 Mar 2026] Title:Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification Authors:Yichi Zhang, Nabeel Seedat, Yinpeng Dong, Peng Cui, Jun Zhu, Mihaela van de Schaar View a PDF of the paper titled Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification, by Yichi Zhang and 5 other authors View PDF HTML (experimental) Abstract:As LLM-powered agents have been used for high-stakes decision-making, such as clinical diagnosis, it becomes critical to develop reliable verification of their decisions to facilitate trustworthy deployment. Yet, existing verifiers usually underperform owing to a lack of domain knowledge and limited calibration. To address this, we establish GLEAN, an agent verification framework with Guideline-grounded Evidence Accumulation that compiles expert-curated protocols into trajectory-informed, well-calibrated correctness signals. GLEAN evaluates the step-wise alignment with domain guidelines and aggregates multi-guideline ratings into surrogate features, which are accumulated along the trajectory and calibrated into correctness probabilities using Bayesian logistic regression. Moreover, the estimated uncertainty triggers active verification, which selectively collects additional evidence for uncertain cases via expanding guideline coverage and performing differential checks. We empirically validate GLEAN with agentic clinical diagnosis acros...