Behavioral Risk Scoring
Overview of the Risk Scoring System
The ProctorAI platform utilizes a dynamic Behavioral Risk Scoring engine to evaluate student integrity in real-time. Each student is assigned a live score ranging from 0 to 100, where 0 represents perfect compliance and 100 indicates critical integrity concerns.
This score is not static; it fluctuates based on a continuous stream of telemetry data processed from the student's webcam, microphone, and browser environment. Proctors can monitor these scores globally via the dashboard to prioritize which sessions require immediate intervention.
Core Risk Indicators
The scoring engine aggregates data from three primary monitoring vectors:
1. Visual Behavior Tracking
Using advanced computer vision simulations, the platform monitors the student's physical presence and focus.
- Gaze Deviation: Detects when a student's eyes move away from the exam interface for sustained periods, suggesting the use of external aids.
- Face Presence & Multi-Face Detection: Triggers alerts if the student leaves the frame or if additional unauthorized persons are detected in the camera's field of view.
- Head Movement Patterns: Identifies unusual or repetitive motion that deviates from standard test-taking behavior.
2. Environmental Audio Intelligence
The system analyzes background audio to distinguish between ambient noise and suspicious activity.
- Audio Anomalies: Detects whispers, vocal cues, or electronic sounds that may indicate communication with external parties.
- Persistent Noise Levels: Visualized via a real-time audio level monitor, identifying environments that fail to meet "quiet zone" requirements.
3. System & Browser Integrity
The platform enforces a "locked" testing environment through browser-level event listeners.
- Tab Switching: Every instance of a student navigating away from the exam tab is logged as a medium-severity violation.
- Fullscreen Violations: The system tracks attempts to exit fullscreen mode, which is often a precursor to accessing other applications.
- Prohibited Interactions: Attempts to use blocked shortcuts (Copy/Paste/Cut) or right-click menus contribute to the cumulative risk score.
Risk Categorization
To help proctors manage large cohorts, risk scores are segmented into four color-coded tiers:
| Category | Score Range | Dashboard Indicator | Action Recommended | | :--- | :--- | :--- | :--- | | Low | 0 – 25 | Green | Normal monitoring; no action required. | | Moderate | 26 – 50 | Yellow | Keep high visibility; potential incidental violations. | | High | 51 – 75 | Orange | Review event timeline; potential intentional cheating. | | Critical | 76 – 100 | Red | Immediate intervention; likely integrity breach. |
Real-Time Analytics & Feedback
The risk score is a live metric that evolves as the session progresses. Proctors interact with this data through several interface elements:
- The Live Alert Feed: A real-time stream of timestamped events (e.g., "Sustained gaze deviation detected") with associated severity levels (Low, Medium, High).
- Risk Trend Visualization: Available in the Analytics view, this area chart displays the average risk score across the entire exam duration, helping institutions identify specific questions or timeframes where violations peaked.
- Violation Distribution: A breakdown of which categories (Gaze, Tab Switch, Audio, etc.) are contributing most to the current risk landscape.
Automated Student Warnings
When specific thresholds are met, the student-facing interface provides immediate feedback:
- Toasts & Notifications: Real-time warnings (via
sonner) inform students when they have exited fullscreen or switched tabs. - Warning Logs: A persistent panel on the student's exam page displays a history of detected anomalies, encouraging self-correction before the risk score reaches a critical level.