(A) The experiment included simultaneous MEG recordings and eye-tracking while participants (N = 24) fixated at all times the fixation cross is present. (B) Acquisition timeline: two 5 min rest sessions (beginning and end), a staircase procedure determining the threshold contrast and a main task involving liminal stimuli presented at a staircase-determined contrast. (C) Each trial of the task included a prestimulus interval, followed by a liminal object stimulus (i.e., leading to ~50% ‘yes’ reports) and two forced-choice decisions: first, ‘category’ (face, house, object, or animal) of the stimulus and, second, ‘recognition’ (yes or no) to indicate whether a meaningful stimulus was perceived or not. A meaningful stimulus was present on most trials (n = 300) whereas a scrambled image was presented on the remaining trials (n = 60, more details in Figure 4). (D) An example 1-min pupil size recording during rest. Light gray trace depicts raw data and dark gray trace depicts the same time course with blink periods excluded and 5 Hz low-pass filter applied. The blue-frame inset shows magnified 10 s recording with pupillary constriction/dilation ‘events’ (red/cyan) and examples of slower ‘states’ spanning 2 s non-overlapping time windows. (E) To study variation in slow states, we defined consecutive non-overlapping 2 s epochs in rest and 2 s baseline periods before each stimulus presentation in task. (F) Power spectrum of a 5 min pupil size recording in rest reveals aperiodic pupil fluctuations since no oscillatory peaks are evident. Transparent light gray curves denote individual subjects and solid black curve denotes across-subject average spectrum. (G) Example distribution of ‘slow state’ pupil size (i.e., averaged in 2 s windows) recorded during rest for one subject. The black lines depict percentiles (20, 40, 60, 80) according to which the 2 s windows were split in groups in Figures 2 and 4. Source data is available as a supplementary file.