How Honorlock Detects Eye Movement During Exams

Honorlock's eye-tracking is one of the more publicized parts of its monitoring. The model watches webcam frames for gaze direction and head pose, flagging when your eyes drift off-screen for more than a few seconds at a time.

Honorlock's eye-tracking pipeline is similar to other AI proctors: extract face landmarks, estimate gaze direction relative to the camera, classify "on-screen" vs "off-screen" per frame. The model has tolerance windows - briefly looking away (2-3 seconds) is normal, repeated long looks away gets flagged. The model does not actually know what you're looking at - just whether your gaze is at the screen. The Fracture overlay sits on the screen, so reading from it counts as "on-screen" gaze. This is why the overlay is well-positioned for AI-proctor environments: it does not require gaze-off-screen patterns.

No - the model classifies gaze relative to the screen, not to specific regions. As long as you look at the screen, you pass.

Reading speed does not directly trigger the eye-tracking model. It triggers other heuristics (typing pattern, response speed) for separate review.

Some users report higher false-positive rates. You can ask the proctor for accommodation; it varies by institution.