Sign In

Phase 4 Live Benchmark

Real telemetry from 10 automation methods posted to /api/score (34-signal behavioral scoring engine). Try the sandbox →

Generated: 2026-03-10T23:28:32Z Sessions/method: 10 Pass threshold: ≥70 Methods tested: 10 Scoring endpoint: /api/score (34 signals)
Method Mean Score 95% CI Min / Max Pass Rate Status
browser_driver ★
80.2
[79.3, 81.1] 77.7 / 82.6 100% live
playwright
0.0
[0.0, 0.0] 0.0 / 0.0 0% live
selenium
0.0
[0.0, 0.0] 0.0 / 0.0 0% live
undetected_chromedriver
0.0
[0.0, 0.0] 0.0 / 0.0 0% live
botright
0.0
[0.0, 0.0] 0.0 / 0.0 0% live
playwright_stealth
0.0
[0.0, 0.0] 0.0 / 0.0 0% live
human — — — — stub
chrome_mcp — — — — stub
puppeteer — — — — N/A
ghost_cursor — — — — N/A

browser_driver uses Fitts’ Law mouse paths, digraph keystroke timing, AR(1) scroll with reading pauses, and OS-level isTrusted=true events. Automation tools (Playwright, Selenium, UDC) score 0 because the detection engine’s event-trust gate fires on isTrusted=false events.

Stub methods (human, chrome_mcp) require external setup. Node.js methods (puppeteer, ghost_cursor) require a separate JS harness. See API docs →