This article explains the algorithm for calculating the Proctoring Credibility Score.
The Proctoring Credibility Score is an automated indicator of a test-taker’s overall integrity during an exam session. It is based on AI-detected behaviors and device activity patterns. Each metric represents a type of potential violation, weighted by its severity or importance.
At this stage, the system does not explicitly detect the use of pen, paper, or notes.
The AI focuses primarily on biometric and behavioral metrics (e.g., face visibility, multiple faces, focus, audio activity).
Such actions can only be confirmed through manual or post-exam review, as part of the Record & Review procedure.
Metrics
For each proctoring session, the system can calculate the average percentage violation for the following metrics:
- Focus changed to a different window (b1)
- Full-screen mode is disabled (b2)
- Webcam is disabled (c1)
- Face invisible or not looking into the camera (c2)
- Atypical keyboard handwriting (k1)
- Microphone muted or its volume is low (m1)
- Conversation or noise in the background (m2)
- No network connection (n1)
- Screen activities are not shared (s1)
- Several faces in front of the camera (c3)
- Face does not match the profile (c4)
- Found a similar profile (c5)
- No connection to a mobile camera (n2)
For easier reference, the metrics are grouped as follows:
- Browser/Screen behavior: b1, b2, s1
- Camera/Face behavior: c1–c6
- Audio behavior: m1, m2
- Network connection: n1, n2
- Keyboard behavior: k1
Detection and Reliability Notes
The list of metrics considered for the final credibility score is defined on the proctoring session template level.
On the same level, each metric can have a weighting coefficient assigned (default weight is 1). This allows fine-tuning of the influence of each metric on the final result.
Please note that the accuracy of AI detections may vary depending on lighting, camera stability, or microphone quality.
Short or low-weight incidents (for example, brief background noise or quick face movement) may be logged but will not appear as red flags unless they lower the credibility score below the defined threshold.
Formula
The credibility score for the proctoring session is calculated using the following formula:
where E ∈ [0, 100] is the credibility score (if E < 0, then E = 0), xk is the session-averaged value of the metric k, wk is the weighting coefficient of the metric k, M ∈ {b1,b2,c1,c2,…} are the metrics.
The system automatically grades each proctoring session based on the Credibility Score (E) and the Threshold (T) parameter, if the auto conclusion is enabled.
The threshold T is defined on the proctoring session template level.
- If E < T, the system automatically assigns a negative conclusion to the proctoring session.
- Otherwise, it assigns a positive conclusion.
The default threshold can be customized per template, allowing institutions to adjust strictness based on their policy.
How it works
- For each violation, the system measures how long the violation lasted during the session and converts it into a percentage of the total session time.
- Each violation percentage is multiplied by its assigned weight.
- All weighted violations are summed.
- This sum is subtracted from 100 to produce the credibility score (in %).
Note: Violations with higher weights have a greater impact on the credibility score. Using very high weights can cause the score to drop significantly, even if the violation occurred for only a short period of time.
Simple Example
Suppose we have a session that lasts 60 minutes. The violations and the percentage of the session during which they occurred can be calculated as follows:
- b1 violation: happened for 12 minutes: b1=12/60⋅100%=20%
- b2 violation: happened for 6 minutes: b2=6/60⋅100%=10%
- c1 violation: happened for 3 minutes: c1=3/60⋅100%=5%
The weights assigned to these violations are:
- wb1 = 0.5
- wb2 = 2
- wc1 = 1
Then, the credibility score is calculated as follows:
100−(wb1⋅b1)−(wb2⋅b2)−(wc1⋅c1)= 100− 0.5⋅20− 2⋅10− 1⋅5 = 100−10−20−5 = 65%
Conclusion: The higher the weight of a violation, the greater its impact on the credibility report.
Interpretation Guidelines
To assist with review and interpretation:
-
85–100% → Minor or no irregularities
-
60–85% → Moderate irregularities (review recommended)
-
Below 60% → Significant irregularities requiring detailed review
