Tablet-Camera Iris Landmarks and EEG/PPG-Derived Arousal Indices During Naturalistic Reading: Geometric Confounding and Exploratory Within-Subject Associations

Arousal, the activation dimension of the valence–arousal circumplex, indexes alertness and engagement independently of pleasantness. We evaluated whether MediaPipe FaceLandmarker iris landmark distances from a tablet camera estimate electroencephalography (EEG)- and photoplethysmography (PPG)-derived arousal markers during naturalistic reading. Recordings from 48 participants yielded 43,923 thirty-second windows. Raw iris distances correlated with EEG alpha (r=+0.118) and theta (r=+0.076), but face scale explained 93% of their variance. After face-width normalisation, alpha remained associated (r=+0.076), whereas theta was not statistically significant. Further adjustment for ocular luminance and head pose rendered alpha nonsignificant (r=+0.021, p=0.33). Circular-shift testing identified alpha coupling in 11/48 participants under full control, versus 28/48 using uncorrected Pearson tests. Within-subject Ridge regression, with no raw sample shared between the training and test windows, yielded a nonsignificant group mean correlation (r=+0.078, p=0.07); 13/48 participants had individually significant prediction–reference associations (11 positive), fewer when the PPG term was restricted to filter-consistent beats. Both counts are exploratory, nominal and uncorrected across participants, and come from different analyses. Neither establishes prospective monitoring performance, and the study does not validate the landmark distance as a pupil measure.