Check data before model training or scoring
Establish provenance, coverage, leakage and operating-domain limits.
Before using this guide
Confirm your site, role and source revision. Follow the local procedure and responsible engineer for physical actions.
Readiness checklist
- Identify source systems, site, equipment and time span.
- Measure missing, stale, simulated and out-of-range data.
- Check label origin and possible future-information leakage.
- Split training and evaluation data in a way that reflects real deployment.
- Record units, calibration changes, operating regimes and model version.
After deployment
Compare observations with the declared domain, monitor drift and use shadow comparison before any operational reliance. Human owners must decide when to stop or retrain.
Source references
These official references provide additional context. Confirm the current edition and your site requirements.