How does look-away detection work, and how far can you trust it?
Look-away detection estimates the direction of the head, or of the eyes, in each camera image and flags images where it points away from the screen. Some tools use gaze, some only head turn.
It is a weak signal on its own. People look away to think, to read a question twice or because a noise startled them. And the software sees only the images it samples, so a glance between two photos is invisible. A turned head tells you where the head pointed, not what the candidate saw.
In Recagn
Recagn doesn't track eyes. With webcam proctoring on, it reads how far the head is turned to the side in each webcam still, after the interview, when it is scored. A run of consecutive stills with the head turned far enough to the side is a Some concern on Integrity. It is never Strong on its own.
Recagn sees only the stills. Snapshots take one every 20 to 30 seconds and Continuous every 8 to 10 seconds, so it can't see a look away between two stills. Leaving the page is a separate rule, read from Activity Tracking.
The finding sits on the timeline at its still and is reported with the other findings as No, Some or Strong concerns, never a number. Recagn's rules list what happened; you decide what it means. It never enters the score or the ranking, and it never stops an interview. The exact turn and run length are on the Integrity rule table.