# 9. Governance
<a name="governance"></a>

ML governance encompasses a set of processes and frameworks that help in the deployment of ML models. It includes model explainability, auditability, traceability, and other more abstract but essential requirements of a successful end-to-end ML lifecycle.


|  |  | 
| --- |--- |
| **9.1 Data quality and compliance** | The ML system accounts for personal identifying information (PII) considerations, including anonymization. It has documented and reviewed column-level lineage for understanding the source, quality, and appropriateness of the data. It also has automated data quality checks for anomalies. | 
| **9.2 Audit and documentation** | The ML system has a full log of all changes during development, including experiments run and reasons for choices made for regulatory compliance. | 
| **9.3 Reproducibility and traceability** | The ML system includes a full data snapshot for precise and rapid model re-instantiation, or it has the ability to recreate the environment and retrain with a data sample. | 
| **9.4 Human-in-the-loop signoff** | The ML system has manual verification and authorization for regulatory compliance. The system requires signoffs for every environment move (for example, Dev, QA, pre-Prod, and Prod). | 
| **9.5 Bias and adversarial attacks testing** | The ML system has *Red Team* adversarial testing using multiple tools and attack vectors, and automated bias checking on specific subpopulations. This component ties back to the Observability and model management section. |