How we detect cheating

Testevy does not guess. Every decision is backed by visual, behavioral, and system-level evidence. Our goal is to protect honest candidates while reliably identifying real fraud.

Action model

Each candidate receives a suspicion score based on all collected signals. We never auto-reject without evidence.

Low risk

Test accepted normally

Medium risk

Flagged for human review

High risk

Confirmed cheating

1. Visual signals

Multiple cameras and AI vision models analyze a candidate’s physical behavior to detect impersonation, device usage, and off-screen assistance.

  • Multiple person detection
  • Face absence or prolonged occlusion
  • Identity consistency checks
  • Suspicious object detection
  • Dual side-camera monitoring to remove blind spots
  • Abnormal gaze persistence away from test screen
  • Repeated micro-glances to consistent off-screen locations
  • Head-eye mismatch detection
  • Cross-view contradiction detection between cameras

2. Typing signals

Human problem solving leaves behavioral fingerprints. We analyze typing dynamics to detect external assistance or automation.

  • Keystroke rhythm analysis for unnatural typing bursts
  • Inactivity-to-completion jumps
  • Copy-like behavior without clipboard usage

3. Browser & system signals

We monitor how candidates interact with their device to detect hidden browsing, tool usage, or screen manipulation.

  • Tab switching and focus loss detection
  • Browser change events
  • Split-screen and multi-window usage
  • Fullscreen enforcement violations
  • Developer tools and console access
  • Abnormal refresh, navigation, and URL changes
  • Cross-signal correlation with gaze and typing