The structured audit framework for every ML pipeline.
A model hit 11.5% weighted MAPE. The team celebrated. A feature audit then found three variables were leaking the target. After removing them, error jumped to 71.3%. The model had never been forecasting — it had been reading the answer sheet.
Audit.md exists so that never happens to you.
- 14 pipeline stages — from problem definition to post-deployment monitoring
- 5 leakage types with detection patterns and code examples
- 100+ audit checks with severity tiers
- Feature Audit Log, Model Card, and Post-Mortem templates
- Walk-forward validation and drift monitoring code
- Pre-training and post-training master checklists
Drop Audit.md into every ML project repo alongside your README.md. Run it at every stage. No exceptions.
Domain-agnostic — works for classification, regression, forecasting, ranking, and anomaly detection. License: Use freely. Star the repo. Audit relentlessly.