Directional dynamics reveal regimes of human visual examination

Eye movements reveal how humans sample complex visual information, from reading to medical diagnosis. In radiology, experts exhibit distinctive scanpath patterns, yet existing gaze metrics primarily quantify how far and how fast the eyes move, without isolating short timescale variability in viewing direction. While computational approaches characterize fixation sequences and spatial patterns, they do not measure moment to moment directional change during ongoing examination. We introduce directional variability, defined as the variability in angular change between consecutive gaze displacements. Across temporal windows from 10 to 50 ms and across both medical image interpretation ([Formula: see text] radiograph examinations) and general viewing tasks ([Formula: see text] recordings), directional variability exhibits a stable bimodal distribution. In contrast, velocity, acceleration, and step length variability show asymmetric spike plus tail distributions without comparable mode separation. Regimes defined by directional variability show large differences in directional entropy and angular reversal density that persist after controlling for spatial extent. At the study level, directional regime proportions improve prediction of scanpath topology metrics and time to report beyond comprehensive magnitude based baselines including velocity, acceleration, and turn statistics. By identifying stable directional regimes across clinical and nonclinical contexts, directional variability represents a domain general geometric marker of examination dynamics with potential applications in expertise research, visual search modeling, human AI interaction, and performance assessment.