experiment-audit

To audit machine learning experiment results for fraud, you can run a cross-model review process that checks for fake ground truth, score normalization fraud, and phantom results to ensure experiment integrity.

Cross-model review is a verification process that uses a secondary model like GPT-5.5 to independently validate machine learning experiment results and detect common fraudulent practices or scope issues.

You can detect fake ground truth or score normalization fraud by running an automated experiment audit that collects your result files and generates a report flagging these specific fraud patterns.

Yes, you can automate the collection and review of machine learning experiment files by running an audit skill that gathers the files, performs a cross-model review, and outputs an audit report.

The experiment audit process does not require any specific external dependencies, allowing you to directly verify your machine learning experiment results without additional environment setup.

Automated audits for machine learning experiments focus on detecting known fraud patterns like fake ground truth and phantom results, but may not catch novel or complex manipulation tactics outside this scope.