Eye-tracking data is widely used to study human attention and behavior, yet raw gaze and pupil signals are noisy, incomplete, and highly variable [2, 3]. Preprocessing is therefore critical, but current workflows often lack standardization and rely on ad-hoc decisions. Small choices—such as imputation, outlier handling, or normalization—can significantly affect downstream analysis and model performance. Existing tools provide isolated functions but little guidance on how to combine them into coherent pipelines. These challenges become particularly relevant when preprocessing decisions must be applied consistently across multiple datasets or repeated analysis runs. CIRRUS therefore focuses on transparent offline preprocessing and reproducible pipeline selection.