Scene 1: Eye-Movement Stability Contrast
Fused streams only (no normalization in the simulator): pupil = tracking-area x/y; gaze = unit direction x/y. Missing samples are exported as null and the trajectory breaks there.
An interactive demo driven by real HPO-CLD Tobii eye-tracking data and LSTM V2.1 warning outputs. Instead of showing static model scores, it lets the audience see how the system detects rising cognitive load from eye-movement instability and turns that signal into human-centered support.
Fused streams only (no normalization in the simulator): pupil = tracking-area x/y; gaze = unit direction x/y. Missing samples are exported as null and the trajectory breaks there.
Adjust the indicator and threshold, then watch how the warning probability (red) co-moves with normalized instability signals (overlays).
The warning does not make decisions for the user. It makes an invisible cognitive state actionable. The audience chooses an intervention and sees how the system can reduce load while keeping the human in control.