What Proctoring Actually Detects

Remote proctoring is sold as catching cheating. What the software actually produces is a list of moments that differed from what it expected. A useful contrast is Monitask: ordinary workplace activity data also needs a declared purpose and human interpretation, so an exam signal should never be treated as self-explanatory.

What the signals are

Face absent from frame, or a second face present.

Gaze directed away from the screen for longer than a threshold.

Sound above a threshold, or speech detected.

Window focus lost, or a second screen connected.

Typing rhythm changing, in systems that model it.

Each is a measurable deviation. None of them is cheating, and each has ordinary explanations.

The gap

Looking away for ten seconds is consistent with reading a hidden note and with thinking.

A voice in the room is consistent with a collaborator and with a flatmate on the phone.

A face leaving the frame is consistent with retrieving material and with a child arriving.

The system cannot distinguish these and does not claim to. It reports the deviation and leaves the interpretation to somebody else.

Which means the product being bought is a list of moments requiring human judgement, not an answer.

What the vendor is actually selling

A reduction in the number of sessions a human has to watch.

That is a real and useful thing: two thousand exams cannot be reviewed in full, and a system that surfaces forty moments makes review possible.

The claim to notice is not detection but triage, and understood that way the technology is far more defensible than the marketing makes it.

What it does not detect

Anything done outside the camera's view with no sound and no device. A note on the wall behind the laptop, a person signalling silently.

Anything done before the exam: obtaining the paper, memorising answers, having somebody else prepare material.

Anything done by a substitute who passed the identity check.

And any form of assistance that does not produce a deviation, which is most of the sophisticated ones.

The asymmetry

It is best at catching the careless and worst at catching the determined.

Which inverts the usual justification: the students most likely to be flagged are those whose circumstances produce deviations, and the students most likely to succeed at cheating are those who prepared for the system.

This is not an argument against using it. It is an argument against claiming it does something else.

What this means for an institution

Buy it for triage, describe it as triage, and design the human step properly — which is the subject of the process section and is where most implementations fail.

Do not tell students it detects cheating, because it does not, and the first appeal will establish that publicly.

What to check

Can your institution state, in one sentence, what the system reports?

Does any of your student-facing material claim it detects cheating?

Who looks at the flagged moments, and what are they permitted to conclude?

And what proportion of flags turn out to be nothing — do you know?

The point

Buy it for triage, describe it as triage, and design the human step properly.

An institution that tells students the system detects cheating will have that claim tested at the first appeal, and it will not survive.

What the vendor is selling

A reduction in the number of sessions a human has to watch.

That is real and useful: two thousand exams cannot be reviewed in full, and a system surfacing forty moments makes review possible. Understood as triage rather than detection, the technology is far more defensible than its marketing makes it.

Worth stating

A reduction in the number of sessions a human has to watch is a real and useful thing to buy.

Two thousand exams cannot be reviewed in full, and a system surfacing forty moments makes review possible — understood as triage, the technology is far more defensible than its marketing.

Also worth knowing

It is best at catching the careless and worst at catching the determined, which inverts the usual justification.

The students most likely to be flagged are those whose circumstances produce deviations; the students most likely to succeed at cheating are those who prepared for the system. For wider institutional context, consult UNESCO.