8 detectors, 10,000 essays: the full benchmark
Balanced accuracy, false positives, and degradation under evasion.
Every study here is built on hands-on testing with documented, reproducible methodology. The data is open so others can verify - or challenge - the findings.
Balanced accuracy, false positives, and degradation under evasion.
Which evasion tools actually defeat detection - and at what quality cost.
A 6.9× disparity that aggregate accuracy hides.
Designing a human-in-the-loop integrity workflow for a large public university.
Standing up provenance checks and a deepfake response plan in six weeks.
An evidence-based procurement process for detection at scale.
Every study follows a documented, reproducible protocol. Read how I source, test, and verify before drawing any conclusion.