Configuration management best practices demand there should be full traceability end-to-end from source code, build, release (model building & model deployment), as well as artifacts (model, container, pipeline runs etc.)

At this time there are gaps in achieving this, especially if concurrent builds are run, e.g.:

• the model deployment pipeline picks up the latest build
• the evaluate_model.py script assumes as described a comment: "Get most recently registered model, we assume that is the model in production"
• When traceability is possible, things are quite buried and require a lot of scripting to surface