Keystroke and Behaviour Analysis

Some systems build a profile of typing rhythm, mouse movement or interaction pattern, and flag departures from it. Behavioural patterns should be handled as cautiously as the records described in the product explanation, with visible criteria and a route to correct misclassification.

What is claimed

That typing rhythm is individual enough to verify identity continuously.

That a change mid-exam indicates a substitution or dictation.

And that interaction patterns distinguish somebody working from somebody transcribing.

What is supported

Typing dynamics do vary between people, measurably, under controlled conditions.

Under exam conditions on unfamiliar keyboards with stress, fatigue and varying question types, the variation within a person approaches the variation between people.

Which is the finding that matters and is rarely mentioned: the signal degrades exactly where it would be used.

Published independent evaluation in assessment settings is thin.

Why the baseline is the problem

A profile has to be built from earlier typing — a practice session, an enrolment exercise, previous coursework.

That baseline is short, taken in different conditions, and frequently on a different device.

A student who typed their baseline on a laptop and the exam on a borrowed desktop will deviate, and the deviation means nothing.

Who it misreads

Anyone using assistive input: speech recognition, switch access, on-screen keyboards, predictive text.

Anyone with a condition affecting motor control, including intermittent ones.

Anyone typing in a second language, where rhythm differs by content.

And anyone whose circumstances changed between baseline and exam, which is most people over a term.

How to treat a behavioural flag

As the weakest signal available, weaker even than gaze.

Never as a basis for a case on its own.

And with an explicit question: what is the student's access arrangement, because assistive technology is the single most common innocent explanation and is knowable in advance.

The question to ask a supplier

What is the false positive rate for students using assistive input?

What baseline length is required, and under what conditions?

Has the model been evaluated by anybody other than the vendor?

A supplier without answers is selling a feature rather than a measurement, and it should be switched off rather than tuned.

The honest use

As one input into an aggregate risk score used purely for triage ordering.

Not surfaced to reviewers as a standalone reason.

And not described to students as identity verification, which overstates it considerably.

What to check

Is behavioural analysis switched on in your configuration — and did anybody decide that?

Can a reviewer see that a flag was behavioural rather than visual?

Do you know which students use assistive input, and is that information available to the review process?

And could you explain a behavioural flag to an appeals panel?

The point

Under exam conditions the variation within a person approaches the variation between people, which is where typing biometrics would be used.

High legal bar, low evidential value, and it misreads assistive input.

Worth stating

Treat a behavioural flag as the weakest signal available, weaker even than gaze, and never as a basis for a case on its own.

Ask explicitly what the student's access arrangement is, because assistive input is the commonest innocent explanation and is knowable in advance.

Also worth knowing

The baseline is the problem: it is short, taken in different conditions, and frequently on a different device.

A student who typed their baseline on a laptop and the exam on a borrowed desktop will deviate, and the deviation means nothing.

And finally

Used honestly, behavioural signals are one input into an aggregate ordering for triage — not surfaced to reviewers as a standalone reason, and not described to students as identity verification, which overstates it considerably..

Summary

Switch it off. High legal bar, weak evidence, and it systematically misreads students using assistive input.

In summary

Behavioural biometrics carries the highest legal bar and the lowest evidential value of anything in this toolkit, and it misreads assistive input systematically.

Switching it off is a straightforward decision. For wider institutional context, consult Jisc.