Jordan Kulwicki¹, Shawn Akridge², Matthew Peterson²; ¹George Mason University

Understanding how visual attention and arousal scale with task demands is central to model of perceptual expertise. Rhythm action video games provide a naturalistic yet controlled environment for testing how visual strategies adapt to increasing task challenge. Here we present a secondary analysis of an existing Clone Hero dataset, collected by Akridge & Peterson (George Mason University), focusing on how experience level modulates confidence, workload, performance, and pupillary arousal under varying speed demands. In the parent study, twenty-four participants played repeated trials at base speed (9) and fast speed (18) while eye movements and pupil size were recorded with an EyeLink 1000, Participants spanned novice, moderate, and expert experience levels. The dataset includes visual attention metrics, NASA-TLX workload subscales, confidence ratings, performance accuracy, and pupil area. We analyzed these measures using ANOVAs and linear mixed-effects models, with an emphasis on expertise and speed related changes in arousal. Replicating core findings, experts showed reduced fixation dispersion, higher accuracy (p <.001), and higher confidence (p <.001) than less experienced players. Speed increased subjective workload across groups, but workload did not differ by experience. Critically, overall pupil size did not differ across experience levels; instead, a significant Speed x Experience interaction emerged, F (2, 222) = 3.36, p = .037. Mixed effects modeling revealed that only experts exhibited reliable increases in pupil dilation with faster speeds (β = 32.56, p = .016), indicating adaptive, context sensitive arousal modulation rather than globally elevated arousal. These results suggest that expertise is characterized not only by more efficient visual attention strategies, but also by the ability to selectively upregulate arousal when task demands increase, supporting accounts that treat arousal as a learned control signal for tuning visual processing efficiency under uncertainty.