Resources
Required tools
- Python with a locked environment
- scikit-learn-compatible baseline pipeline
- data-version fingerprinting
- experiment metadata in machine-readable form
- tests for split integrity and feature availability
Artifact schemas
A run record includes commit, environment, dataset fingerprint, configuration, seed policy, start time, metrics, slices, produced artifacts, and parent run. A decision record includes owner, threshold, capacity assumption, fallback, expiry, and approver.
Primary references
- scikit-learn documentation
- NIST AI RMF
- Datasheets for Datasets
- Model Cards
- Designing Data-Intensive Applications
Prefer specifications and primary papers for claims. Treat blog posts as leads, not authority.