Triggers when deploying a trained model to production — packaging, batching, GPU/CPU sizing, versioning, shadow/canary, latency budgets, and monitoring drift and inputs.
