Probiotic supplementation promotes bone development, immune maturation, and cognitive function in growing mice.

Decision Support Systems (DSS) are increasingly implemented as hybrid pipelines in which machine learning models generate recommendations and humans decide whether to follow or override them. In this setting, system-level performance depends on interaction dynamics rather than model accuracy alone. However, most publicly used benchmarks remain prediction-centric and provide limited support for evaluating human–AI decision behavior. This benchmark gap is now a technical bottleneck. In operational environments such as healthcare, cybersecurity, and IoT management, practitioners must decide under uncertainty, time pressure, and incomplete information. These conditions directly affect trust, reliance, and override frequency, but they are rarely encoded in standard datasets. As a result, many methods can be compared on predictive metrics, while few can be compared on interaction-aware decision quality. Prior work in human-in-the-loop and interactive machine learning has shown that human intervention is heterogeneous: domain experts may correct model failures, whereas low-expertise users may introduce additional errors through unnecessary overrides. Yet reproducible, decision-level datasets that jointly represent context, recommendation confidence, explanation signals, user action, and final outcome remain scarce.