DEMM-Bench: A Cross-Regime Benchmark for Agent-Runtime Governance-Evidence Sufficiency

DEMM-Bench is a newly introduced benchmark aimed at assessing the sufficiency of governance evidence in agent-runtime systems, based on the Decision Evidence Maturity Model (DEMM). It evaluates records across eight evidence regimes to determine their ability to reconstruct decision-level properties, revealing that existing baselines often overstate their sufficiency. This benchmark, which includes a dataset of 64 cases and various evaluation tools, is significant for practitioners as it provides a standardized method for assessing the maturity of decision-making evidence, enhancing the reliability of governance in AI systems.