Not every online exam needs the same level of supervision. A weekly knowledge check, an open-book case analysis, and a professional certification exam may run on the same platform, but they serve different purposes, create different integrity risks, and call for different controls.
The better question is not how closely every candidate should be watched. It is what each assessment must prove. Once educators define that, they can choose a proportionate mix of task design, identity checks, access controls, observation and review, guided by shared standards rather than a single supervision rule.
Assessment Purpose Sets the Level of Assurance
A low-stakes quiz may be most useful as a timely indication of current understanding. Its value often lies in helping educators identify what needs further teaching. A professional certification exam may support entry into a regulated occupation, so identity assurance, controlled conditions, and defensible records carry much greater weight.
Format matters as well. An open book task accepts access to information but tests how candidates interpret and apply it. An essay may require evidence of authorship and sustained reasoning. A timed test may be vulnerable to answer sharing or outside assistance. A practical assessment must show whether a candidate can perform a skill, which may require direct observation, a recorded performance or an oral explanation.
This gives assessment teams a clear first step: state what the result is intended to prove, identify what could invalidate that claim, and choose controls that address those risks. The level of supervision then follows from the assessment purpose rather than from an institution-wide default.
Assessment Design Creates the First Integrity Control
Supervision can confirm identity, restrict unauthorised help and create evidence for later review. It cannot make a weak task reveal meaningful learning.
A recall-heavy test remains a recall-heavy test under close observation. If an answer can be reproduced without understanding, increasing scrutiny may discourage some misconduct, but it does not improve the validity of the task. Educators gain more assurance by designing questions that require candidates to interpret evidence, justify decisions, connect ideas or demonstrate a process.
This is why assessment reform is becoming inseparable from integrity policy. TEQSA’s 2025 resource, Enacting assessment reform in a time of artificial intelligence, recommends multiple, inclusive and contextualised approaches to forming trustworthy judgements about learning. It also places security at meaningful points within a wider assessment system rather than treating detection or observation as the whole solution.
In practice, that may mean adding a short oral explanation to a written task, collecting evidence at several stages of a project, using scenario-based questions that require disciplinary judgement, or combining a practical demonstration with questions about the choices made. Each method gives educators evidence that is harder to separate from the intended learning.
Consistency Comes From a Shared Decision Framework
Assessment directors still need common standards. Candidates should know which resources are permitted, how identity will be confirmed, what conduct may trigger review, and how decisions will be made. Staff need consistent evidence thresholds, escalation processes and responsibilities.
The delivery method, however, does not have to be identical. NSW Department of Education’s Elements of effective assessment place equity, validity, reliability and transparency alongside one another. It defines reliable assessment as producing consistent results without avoidable influence from chance, bias, systematic error or cheating.
A shared framework can therefore require every assessment to record its purpose, stakes, likely integrity risks, permitted resources, candidate access needs and review process. One exam may justify live observation and strict resource restrictions. Another may use identity checks, randomised questions and later review. A practical task may require recorded performance plus an oral explanation.
Consistency lies in the rigour of the reasoning. Educators apply the same questions and evidence standards, while retaining the professional judgement needed to match controls to the task.
Candidate Context Sharpens Proportionate Supervision
Even assessments with similar academic stakes may require different delivery arrangements. Candidates may be spread across countries and time zones. Some may have limited bandwidth, shared living spaces, or assistive technology. A cohort of fifty creates different staffing and scheduling demands from a cohort of fifty thousand.
These conditions shape whether a supervision model can operate fairly and whether unusual behaviour can be interpreted accurately. They are therefore part of integrity design, not an adjustment made after the security model has been selected.
The 2025 Open Praxis review of online proctoring in open and distance learning discusses strategies such as multiple forms of authentication, asynchronous review and open book assessment, while emphasising the need to consider accessibility, inclusion and student wellbeing alongside security.
For educators, this supports a practical sequence. Check whether the required technology is dependable for the actual cohort. Confirm that identity and room requirements can be met with approved adjustments. Make permitted resources explicit. Test how assistive technology interacts with the assessment environment. These steps reduce avoidable interruptions and give reviewers better context when an alert or incident is examined.
Layered Controls Connect Risks to Evidence
A proportionate model treats integrity as several connected layers. Task design determines what evidence will be produced. Identity controls establish who is completing the task. Access controls define which people, materials and digital resources are permitted. Observation records relevant behaviour. Review procedures determine how signals are interpreted.
Each risk can then be matched to the control most likely to address it. Impersonation calls for identity assurance. Answer sharing may require question versioning, item exposure controls or timing changes. Unauthorised assistance may be addressed through task design, clear resource rules and proportionate observation. Doubts about authorship may be resolved through process evidence or a brief oral follow-up.
This approach avoids asking one control to solve every problem. It also makes the rationale easier for educators to explain, because each measure has a defined purpose and a clear relationship to the evidence needed from the assessment.
The Choice Belongs Inside Assessment Design
The practical question is not whether one supervision method is universally better. It is which combination of identity checks, observation, task design, access controls and review procedures provides enough assurance for this particular result.
A comparison of remote proctoring vs. invigilators can support that wider decision by clarifying what each approach makes possible, what it requires operationally, and where human judgement enters the process. It should not replace the prior questions about purpose, stakes, format, and candidate context.
A useful decision record explains what the assessment is intended to prove, how misconduct could undermine that claim, which controls address each risk, and what burden those controls place on candidates and staff. It also identifies who reviews alerts or incident reports, what evidence is required and how candidates can respond.
Observation produces signals, not automatic verdicts. Clear review thresholds and access to relevant records help educators distinguish meaningful evidence from behaviour with an ordinary explanation. Where several reviewers are involved, shared examples and moderation can support more consistent judgement.
Review Keeps the Model Aligned with the Assessment
A supervision model should be reconsidered whenever the assessment changes substantially. A new task format, larger cohort, different candidate location, revised technology or higher consequence may alter the risk profile. The control chosen for an earlier version of an exam should not remain simply because it is familiar.
Assessment teams can review evidence that reflects both assurance and delivery. Useful indicators include completion rates, technical interruptions, support contacts, the proportion of alerts that lead to further review, case resolution time, appeal outcomes and patterns in approved adjustments. TEQSA’s framework similarly asks institutions what data they collect, what indicators would signal a need for adjustment and how review processes maintain assurance over time.
The purpose is not to create a universal target. It is to identify where instructions, settings, task design or review processes can be improved. Repeated questions about permitted resources may indicate that candidate guidance needs greater clarity. Alerts that rarely produce usable evidence may justify a review of settings or thresholds. Recurring authorship concerns may suggest that staged work or a short oral component would provide stronger evidence than additional observation.
The strongest supervision model is therefore not automatically the strictest or most visible. It is the one educators can explain in relation to the learning evidence the assessment must produce, apply consistently to the relevant cohort, and refine when new evidence becomes available.
