Anti-cheat signals
The behavioral signals the platform analyzes to detect inappropriate help during an assessment.
Anti-cheat signals
Beyond video, the platform analyzes behavior during the test to surface potential cheating. These are the same signals summarized in the integrity report.
A signal is a lead, not a verdict
No single signal proves cheating. The platform presents them with timing so you can correlate with the video and code. The goal is to give you leads to investigate, not automatic judgments.
Paste detection
The system tracks when large blocks of code are pasted rather than typed. A single paste is not a verdict, but a pasted, fully formed solution after the candidate looked away is a stronger signal.
Focus and window events
If the candidate switches tabs or leaves the window, that is recorded with a timestamp. Frequent or long absences during a coding task are worth a closer look.
AI assist detection
Patterns in the code — sudden style shifts, structure inconsistent with the rest of the session, or signs of AI-generated solutions — contribute to the transcription and logical leap signals.
What the platform refuses to do
A few rules are enforced by the platform to keep the bar honest:
- The bar may never reject on a fraud signal. Bulk paste, transcription, phone
detected, tab switch, multiple faces, and LLM-assisted are all leads. The
the platform refuses
DISQUALIFYfor these flags when a policy is written. - Automation never decides an integrity concern. Even when the policy is allowed to act unattended, the policy refuses to write a verdict that would reject on a fraud signal.
- A missing dimension is not a zero. A quality axis that was never measured is shown as absent, not as a fabricated score.
Putting it together
| Signal type | Example | What to do |
|---|---|---|
| Bulk paste | A full function appears at once | Check the video around the timestamp; legitimate pastes from documentation are common |
| Transcription | Typing cadence matches copying | Look for an external screen in the video |
| Logical leap | Sudden appearance of a complex solution | Cross-check with the rest of the session |
| Tab switch | Candidate left the tab during a hard problem | Confirm with the video; many candidates look up APIs |
| Phone detected | A phone was visible | Confirm with the video; glancing at a phone is different from reading from it |
| Multiple faces | More than one person visible | Confirm with the video; some environments have mirrors or reflections |
| LLM-assisted | Solution structure consistent with model output | Cross-check with the rest of the session |
Avoid false positives
Good candidates research and reference documentation. Judge signals in context using the synced timeline before acting on a flag.