GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification

Yichi Zhang, Nabeel Seedat, Yinpeng Dong, Peng Cui, Jun Zhu, Mihaela Van Der Schaar
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:158266-158287, 2026.

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 across three diseases from the MIMIC-IV dataset, surpassing the best baseline by 12% in AUROC and 50% in Brier score reduction, which confirms the effectiveness in both discrimination and calibration. In addition, the expert study with clinicians recognizes GLEAN’s utility in practice.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-zhang26fc, title = {{GLEAN}: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification}, author = {Zhang, Yichi and Seedat, Nabeel and Dong, Yinpeng and Cui, Peng and Zhu, Jun and Van Der Schaar, Mihaela}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {158266--158287}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhang26fc/zhang26fc.pdf}, url = {https://proceedings.mlr.press/v306/zhang26fc.html}, 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 across three diseases from the MIMIC-IV dataset, surpassing the best baseline by 12% in AUROC and 50% in Brier score reduction, which confirms the effectiveness in both discrimination and calibration. In addition, the expert study with clinicians recognizes GLEAN’s utility in practice.} }
Endnote
%0 Conference Paper %T GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification %A Yichi Zhang %A Nabeel Seedat %A Yinpeng Dong %A Peng Cui %A Jun Zhu %A Mihaela Van Der Schaar %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhang26fc %I PMLR %P 158266--158287 %U https://proceedings.mlr.press/v306/zhang26fc.html %V 306 %X 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 across three diseases from the MIMIC-IV dataset, surpassing the best baseline by 12% in AUROC and 50% in Brier score reduction, which confirms the effectiveness in both discrimination and calibration. In addition, the expert study with clinicians recognizes GLEAN’s utility in practice.
APA
Zhang, Y., Seedat, N., Dong, Y., Cui, P., Zhu, J. & Van Der Schaar, M.. (2026). GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:158266-158287 Available from https://proceedings.mlr.press/v306/zhang26fc.html.

Related Material