Intelligent nuclear power control rooms require automation responsive to task context and operator state. A repeated-measures simulation tested four stage-based levels of automation (LOAs) in regular-monitoring and simplified emergency diagnosis-and-response tasks. Sixteen engineering participants without operating experience completed all conditions; eight with nuclear-engineering backgrounds provided descriptive context. LOA effects varied by task and outcome. Subjective workload followed a quadratic pattern during regular monitoring but generally decreased in the emergency task, whereas reaction time showed a nonlinear, task-dependent pattern. Raw fixation counts were influenced by Time of Interest duration. After normalisation, fixation rate decreased linearly across LOAs only in regular monitoring and was unrelated to subjective workload. Pupil diameter did not consistently track workload. These exploratory findings show that automation effects depend on task definition, outcome measure, and duration normalisation, supporting context-specific evaluation of workload, visual attention, and performance.
