An auditor asks which data and code produced the model that was in production on 3 March. What do you need to have kept?
Options
- Only the final accuracy number
- A screenshot of the dashboard
- A registry entry linking each model version to its code commit, data snapshot, parameters, evaluation results, and the deployment history (Correct option)
- The laptop of the data scientist
Answer
A registry entry linking each model version to its code commit, data snapshot, parameters, evaluation results, and the deployment history
Why
Lineage is the traceable chain from the production model back to its sources. Without it, "which model was live on 3 March?" and "what was it trained on?" cannot be answered.
Keeping this record automatically in the pipeline and the registry is far more reliable than trying to rebuild it afterwards.
It also helps in incidents, to know which other models used the same data.
Remember
Keep lineage: model to code, data and parameters.