A deterministic benchmark for browser-use, Computer Use, and Operator-style agents. 30+ signals across event trust, trajectory, environment fingerprint, and rendering surfaces. Your agent gets a score and a leak list — before any anti-fraud layer ever sees it.
Catch agent tells before any anti-fraud layer does.
A/B different agent strategies and measure how human each one looks. The full check list is deterministic — re-runs against the same headless config produce the same score, so you can actually compare runs.
Comprehensive intelligence on your browser agent — fingerprint, network, and behavior.
Canvas hashes, WebGL renderer, audio context sum, font enumeration, hardware concurrency. Each surface scored against generic / leak / unique.
Visitor IP, ASN, geo, ISP. Cross-checked against navigator timezone and primary language. Disagreement = proxy leak.
Pointer trajectory shape, velocity jitter, click dwell, isTrusted ratio, movementX/Y consistency. Rule weights are visible in source.
Move through the targets. Click each. Drag the last. We score eight signals live.
Worker re-fingerprint · descriptor lies · extra canvas tests · stable hash · surfaces every public auditor exposes.
| Surface | Main | Worker |
|---|
…first visit1Every check is deterministic. Every weight is visible. No black box.
looking up…