Understanding how workload shapes error occurrence remains a central challenge in safety research, particularly in complex service systems characterized by dynamic workflows and incomplete error reporting. From a Safety-I perspective, heavier workloads are viewed as a contributor to increased errors, whereas Safety-II emphasizes how frontline workers adapt processes to sustain performance under pressure. We examine these competing perspectives using two years of high-resolution data from an outpatient cancer facility, combining Real-Time Location System (RTLS) traces, appointment schedules, and reported safety events. Analyzing distinct safety-related outcomes, we identify a nonlinear relationship between workload and performance: moderate workload levels are associated with higher outcome levels, whereas very high workload levels correspond to improved performance, consistent with adaptive responses that stabilize workflow under pressure. Mediation analyses provide evidence for a specific adaptive mechanism − earlier initiation of daily treatments − which improves schedule adherence and stabilizes downstream flow. By integrating reported events with RTLS-based measures of operational disruption, the study enables differentiation between genuine changes in outcomes and effects driven by reporting behavior. Overall, the study makes visible how adaptive behavior shapes performance under high workload and demonstrates how sensor-based data can be used to capture work-as-done. In doing so, it contributes to safety theory by providing empirical evidence on adaptive mechanisms and by opening new avenues for studying adaptive behavior in safety–critical environments.