Proctoring flags: signals, not verdicts
Proctoring flags on NUADU are objective events, not findings: a named invigilator reviews each after the fact, the institution decides, candidates can contest.
Proctoring flags are events logged during an online exam for a person to review — a tab switch, an identity mismatch, an unusual pattern — and not findings that a candidate cheated. On NUADU, a named invigilator reviews each one after the fact, the institution's academic-integrity process decides, the candidate can contest, and the system penalises nobody. This post follows one illustrative flag from event to decision, for the head of assessment who owns the process and the exam officer who runs the session.
What are proctoring flags?
They are the output of the controls an institution chose for an assessment, and on NUADU each one records what happened, not what it meant. The Conduct page gives three examples:
- Browser controls: tab-switching detection, application detection and copy-paste monitoring.
- Identity verification at exam start: an identity-mismatch flag when the face does not match the enrolled photo.
- An unusual pattern: the third example the Conduct page gives.
What never becomes a flag matters as much. Nothing about what a candidate felt or meant is read from their face, voice, gaze or keystrokes, which keeps the reviewer's job possible: a tab switch either appears in the record or it does not.
How much a tab switch weighs is settled before the exam, not when the flag appears. Severity is set per assessment, which lets a weekly quiz and a final weigh the same event differently, and which controls run at all depends on the rung chosen among the types of online proctoring.
Signals are surfaces — not findings.
What happens if you get flagged on a proctored exam?
On NUADU, nothing happens to the candidate at the moment of the flag. The event goes into the session's audit trail, a named invigilator reviews it after the fact, and what follows, if anything, is for the institution's academic-integrity process to decide. Take one illustrative tab switch in a timed exam running the defaults:
- The event is recorded. The system docks no mark, attaches no label and reaches no conclusion.
- A named invigilator reviews it after the fact, with the recording if recorded review was switched on for that assessment, or with the event record and session trail if it was not.
- The institution decides, through the academic-integrity process it already runs, and candidate due process stays intact.
That is the Conduct page's sequence in full: signals go to a named invigilator, the institution's academic-integrity process decides, and the system does not penalise candidates. The invigilator establishes what the record shows rather than passing sentence, and the invigilation workflow on the Conduct page places that review inside the wider layered model.
Does a proctoring flag mean you cheated?
No. A flag means an event happened; whether it means anything is for a named person to establish and the institution to decide. The review step exists for that gap, because false positives are not evenly spread: the Conduct page names disabled, neurodivergent and non-white candidates as the people on whom false-positive flags fall disproportionately.
Proctoring false positives are also why no false-positive rate is quoted here. A rate would describe the software; the decision about a candidate belongs to people. NUADU does not claim to detect every form of misconduct either. The browser controls cover tab switching, application detection and copy-paste monitoring, identity verification surfaces a mismatch at exam start, and recorded review, where it is on, gives a named invigilator something to look at.
If the institution opens an academic integrity investigation, what NUADU contributes is the event and the invigilator's review of it: a record, not a conclusion waiting to be confirmed. The wider online proctoring privacy concerns behind false flags have their own post.
Cheating detection or integrity signals?
Integrity signals. The first framing assumes the software's job is to catch someone; on NUADU the job is narrower and checkable — record the event, surface it to a named invigilator, leave the finding to the institution. AI never decides a grade, a flag or a credential.
The AI Act puts AI systems meant for monitoring and detecting prohibited student behaviour during tests, in education and training institutions, on its high-risk list (Annex III, point 3(d), of the AI Act), a category the Act surrounds with obligations, human oversight among them. Whatever a given system's classification, a named person reviewing every signal is what keeps the decision explainable.
Can a candidate contest a proctoring flag?
Yes — through the institution's defined process, with the same record available to the appeal. The Certification path lists proctoring reports usable as evidence in appeals proceedings; on NUADU that report is the record of what was raised and who reviewed it, not a verdict the panel is asked to approve. Followed back through the trail, one flag shows:
- The event, as written to the session's audit trail.
- The configuration: which controls were on, at what severity, and whether recording was on.
- The review: the named invigilator who looked at it, after the fact.
- The decision: made by the institution's academic-integrity process, not by the system.
- The contest: the candidate's route to challenge the result, with the same record available to it.
Nothing on that list is an inference about the candidate, and nothing on it was decided by software. How an appeals panel reads that record beside the marking record is the subject of the exam appeals process, and the certification exam path sets out the chain a certifying body has to defend.