# Model deployment and operating evidence

Run `python deployment_cases.py` from the repository root. The lab creates an original synthetic subscription fixture and trains a small scikit-learn pipeline. All identifiers, requests, outcomes and incidents are invented.

The verifier covers strict prediction schemas, batch/single equivalence, model-plus-preprocessing reload, error contracts, hashes, experiment records, offline and online evidence, shadow traffic, canary rollback, input and concept drift, delayed labels, retraining gates, rollback order, a reviewed container specification, local load timing, percentile interpretation, artifact integrity, release readiness and handover ownership.

`Dockerfile.review` and `requirements.review.txt` are instructional specifications. No Docker build or network service is executed or claimed. Local timing varies by machine and is not a production service-level result.