The silent flag problem
In most remote proctoring, the candidate is the last to know. The system quietly records that a face left the frame, that a second person appeared, or that their eyes kept drifting off-screen — and the candidate finds out days later, when the session is already flagged. By then the story is fixed: a list of red marks with no chance to explain or correct.
Yet many of those marks are innocent. A candidate leans out of frame to reach a charger. They glance at a second monitor they forgot to unplug. They look away to think through a hard question. None of this is cheating, but all of it looks like a violation on the reviewer's screen.
A heads-up changes the outcome, not just the feeling
ProctorSafe can now show the candidate a short, non-accusatory banner the moment a recurring issue appears: "We can't see you right now. Please make sure your face stays in the camera frame." The effect is twofold:
- Real problems get fixed in the moment. A candidate who keeps drifting off-screen refocuses; a candidate leaning out of frame leans back. The session that would have been flagged simply isn't.
- Ambiguous behaviour is disambiguated. A candidate who is nudged and corrects is demonstrably acting in good faith — useful signal for a reviewer, and fairer to the candidate.
Crucially, the copy is observable, not accusing. It describes what the system sees ("we can't see you right now"), never what it assumes ("you are cheating"). Wording avoids relying on any single sense, so it reads fairly for every candidate.
Designed to be quiet
A warning system that nags is worse than none at all. ProctorSafe leans deliberately toward few, well-timed messages:
- Each issue only warns after it repeats, respects a cooldown, and is hard-capped per session.
- Momentary blips are ignored entirely.
- Once a session's integrity is already low, warnings stop — there is nothing left to salvage by nagging.
The banner is never modal. It never blocks the exam, disappears on its own, can be dismissed, is announced politely to screen readers, honours reduced-motion preferences, and makes no sound. Warning text is chosen and displayed entirely on the candidate's device — the wording never leaves the browser. Only a lightweight audit event, recording that a warning was shown and why, is sent for review.
Opt-in, and off by default
Because tone and cadence are contextual, the feature ships off. A tenant manager turns it on in the dashboard, picks which issues warrant a warning, edits the wording, and tunes how often each may appear. Nothing changes for existing exams until a manager decides it should.
The goal is not more surveillance — it is turning silent judgement into a fair, in-the-moment conversation, and letting honest candidates stay honest without a scare weeks later.