Every unsupervised GCAT session produces a Confidence Score: a figure from 0 to 100 showing how confident the system is that the test was completed independently, under standard conditions.
It is there to help with the questions any online result raises. Can I trust this score? Were the conditions fair? Should I follow up, or offer a supervised retest? The Confidence Score does not detect cheating and never makes a decision. A low score is a reason to look more closely, not proof that anything went wrong.
What the Score Draws On
The score combines several behavioural signals recorded during the session. We don't publish how they are weighted, because that would make them easier to get around.
Leaving the Test Window
How often the candidate switched away from the assessment, for example to another tab or application.
Response Timing
Time spent on each question compared with what is typical for that question, based on large normative samples.
Copy and Paste Attempts
Clipboard access is blocked during the test, and every attempt to copy or paste is recorded.
Restarts
How many times the candidate restarted or recovered the test before finishing it.
Facial Validation (Optional)
Where you use it, Facial Validation adds evidence about who was present and how settled the test environment was. Interpreting Facial Validation Results
Alongside the score and its level, you will see how many times the candidate left the test window, and whether the questions where that happened showed significant timing differences.
The Five Confidence Levels
Scores are grouped into five levels, so you can see at a glance which results need attention.
Very High
Behaviour typical of independent completion. No anomalies detected.
Suggested Response No action needed.
High
Some variation, within the range seen in ordinary sessions. Likely to be a reliable result.
Suggested Response No action usually needed.
Medium
Moderate anomalies. The result may have been affected.
Suggested Response Review the detail alongside your other evidence.
Low
Significant irregularities. Confidence in the result is reduced.
Suggested Response Review carefully and consider a follow-up conversation.
Very Low
Several concerning patterns together.
Suggested Response Treat the result with caution. A follow-up conversation or supervised retest is advisable.
How to Read a Score
Start With Context
Consider the role, the candidate's background and the rest of your assessment evidence. Someone with a strong numerical background may reasonably move faster or slower through numerical questions.
Count the Window Changes
One or two can be innocent, such as a pop-up notification or a stray click. Frequent changes are more likely to mean distraction, or looking at something else.
Check the Timing on Those Questions
The system shows whether the questions where the candidate left the window took unusually long or short. One odd timing is normal; repeated or extreme differences deserve attention.
Compare With Other Evidence
Does the result fit the candidate's interview, CV and references? A score far above what the rest of their profile suggests is worth a closer look.
Look at the Signals Together
A window change or an unusual timing means little on its own. Combined with a low score and other signals, such as a Facial Validation alert, it calls for more caution.
Three Example Scenarios
Scenario 1
The candidate left the test window three times, and each of those questions shows a large timing difference.
This combination may suggest outside help. Review the detail and consider a follow-up conversation.
Scenario 2
A candidate with a low score tells you their Wi-Fi kept dropping during the test.
Technical problems lower the score too. A conversation or a supervised retest may settle it.
Scenario 3
A candidate has a high test score, but the session shows window changes and timing differences, and the result is out of line with their application.
Be cautious. A follow-up conversation or a retest is advisable before you rely on the result.
A low Confidence Score is a reason to look closer. It is never proof of misconduct.
The browser can confirm that a candidate left the test screen, but not where they went or why. The score flags results for a person to review, and any decision always rests with a trained administrator.
What to Do Next
Use the score to decide where to look, not to pass or fail a candidate on its own.
Flag Results for Review
Give low-scoring results a closer look before you rely on them.
Inform Retest Decisions
If a candidate reports technical problems, or concerns are high, offer a supervised retest.
Guide a Follow-Up Conversation
Talk to the candidate, openly and without accusation, about what happened.
Questions You Might Ask
- Did you have any technical problems during the test? How did they affect you?
- Were you aware that leaving the assessment window is recorded as part of our test security?
- Our records show you left the assessment window several times. Was there a reason you needed to?
- Can you confirm that you completed the assessment on your own, without help from other people or from tools such as AI assistants?
If a Candidate Disputes a Result
- Explain how the Confidence Score works.
- Consider offering a retest under supervised conditions.
- Record what you did and why.
Full details of each event behind a score are stored securely and are available to administrators on request.
Things to Keep in Mind
It Isn't a Cheating Detector
A low score is an alert to possible irregularities, not proof of misconduct.
Technical Problems Lower Scores
A poor connection, a power cut or an accidental window switch are all plausible explanations.
Results Can Still Be Valid
Some candidates do well despite disruption. Others underperform for reasons the system cannot see.
Treat Candidates Fairly
Never accuse a candidate of dishonesty without clear evidence, and raise any concern respectfully.
The Evidence Behind It
In a controlled experiment, the Confidence Score told apart candidates who used outside help from those who didn't, and skill with AI tools did not help candidates avoid it. Field data from more than 106,000 GCAT sessions found scores consistent across gender, age and ethnicity groups.
Read the Integrity Research