Which approach should be used when deploying machine learning models to production?

Correct answer: Version models, implement monitoring, use canary/blue-green deployments, and establish rollback procedures

Option C is correct because production ML deployments require model versioning to track changes, continuous monitoring to detect data drift and performance degradation, canary or blue-green deployment strategies to reduce rollout risk, and rollback procedures to recover quickly if a new version underperforms. Option A is incorrect because using training data for monitoring introduces data leakage and does not reflect real-world production distribution, making it unsuitable for detecting concept drift. Option B is incorrect because ML models degrade over time as data distributions shift, so deploying and never updating leads to deteriorating prediction quality. Option D is incorrect because deploying without testing bypasses validation of model correctness, integration, and performance, which is a critical failure in any production engineering practice.