Coding Tests That ChatGPT Cannot Ace
Traditional coding tests are broken — LLMs ace them in seconds. ClarityHire's code coherence AI scores the edit pattern, not the output, so LLM-generated answers get caught.
Code coherence analysis
Signal, not just output
Evidence per flag
Multi-layered detection
Every signal LLMs can't fake
Edit patterns, keystroke rhythm, paste events, and gaze — LLM-generated code can't simulate all of them.
Code coherence AI
Claude scores the edit pattern of every session. Clean LLM pastes stand out.
Keystroke rhythm analysis
XGBoost classifier detects non-human typing patterns and mid-session takeovers.
Paste event correlation
Paste event size correlated with final code — large clean pastes get flagged.
Iteration analysis
Authentic coding shows iteration. Type-test-refactor patterns score higher than drop-and-ship.
Gaze anomaly detection
Off-screen reading during coding tasks suggests external LLM or reference use.
Tab-switch frequency
Frequent tab switches during coding correlate with external LLM queries.
Full session replay
Replay every keystroke, paste, and run to verify any flag manually.
Per-signal authenticity score
See exactly which signals diverged — no black-box verdicts.
Why teams pick ClarityHire for this
Real workflow, built-in integrity, end-to-end pipeline — in one platform.
Score the process, not the result
LLMs produce correct-looking code instantly. ClarityHire evaluates HOW the code was written — the edit sequence, paste timing, and keystroke rhythm — so LLM-generated answers get caught even when the output is perfect.
- Edit pattern analysis (type → test → refactor)
- Paste size + timing correlation
- Keystroke rhythm biometrics
Every signal, every session
Keystroke biometrics, face continuity, gaze analysis, and code coherence AI run silently throughout every assessment. Every signal feeds a per-candidate authenticity score you can review or automate on.
- Ten signal types including keystroke, gaze, face, paste events
- Per-signal authenticity score per submission
- Human-readable explanations on every flag
One platform, not five tools
Job boards, assessments, live video interviews, candidate tracking, structured feedback, and PDF reports — all in ClarityHire. Stop paying for five vendors.
- Replaces ATS + assessment + interview + proctoring
- Kanban pipelines with stage automations
- PDF hiring reports for stakeholders
How it works
Enable strict integrity
Strict mode turns on all ten signals including code coherence AI.
Candidates code
Every keystroke, paste, and edit captured alongside final code.
AI analysis on submission
Claude analyzes the session trajectory and produces authenticity score + flags.
Review evidence, decide
Flags come with AI explanations and session replay links.
Frequently asked questions
How do you catch ChatGPT-generated code specifically?+
Code coherence AI analyzes the edit pattern. ChatGPT-generated code typically arrives as a large clean paste with consistent style and no iterative debugging. Claude flags those patterns with a natural-language explanation.
What about Copilot and other inline AI tools?+
Inline AI suggestions produce a different signature than bulk paste — rapid repeated accepts across lines. ClarityHire's keystroke rhythm classifier catches this pattern.
Does this produce false positives?+
False positives are always possible with any heuristic. ClarityHire combines code coherence AI with 9 other signals to reduce false flags, and every flag comes with a human-readable explanation so your team can verify before acting.
Can I tune sensitivity per role?+
Yes. Three integrity levels (none / basic / strict) per assessment. Confidence thresholds adjustable for senior candidates who may genuinely code cleanly and quickly.
Is code sent to a third-party AI?+
For code coherence analysis, yes — Anthropic or OpenAI, both vetted data processors with signed DPAs. Self-hosted Ollama option available for regulated orgs where code cannot leave infrastructure.
Explore related features, roles, and use cases
ClarityHire is one platform. Every feature is built to work with the rest.
Code Coherence AI
Claude-powered analysis of edit patterns for AI-proof grading.
Cheat Detection
Ten signal types for AI-proof assessments.
Keystroke Biometrics
XGBoost classifier for typing-rhythm analysis.
Detect ChatGPT Cheating
AI-proof integrity in depth.
Take-Home Coding Tests
Async take-homes with AI-proof integrity.
Run AI-Proof Coding Tests Today
Code coherence AI, keystroke biometrics, and explainable flags. Free to try.