What problem does it solve? Turning notebook experiments into production ML systems often fails due to data leakage, irreproducible training, missing evaluation gates, and absent monitoring or rollback paths. This Skill provides a structured workflow that forces explicit data contracts, reproducible pipelines, measurable promotion criteria, and operational monitoring before a model ships. ## Core Features & Use Cases - Prediction and Data Contracts: Define entity grain, label timing, feature freshness, split policy, and leakage checks before writing model code. - Reproducible Training and Promotion Gates: Pin configs, seeds, and dataset versions, then enforce automated metric gates comparing against baselines and the current production model. - Serving, Monitoring, and Rollback: Package versioned artifacts with schema validation, fallback behavior, drift monitoring, and named rollback triggers. - Use Case: When converting a churn-prediction notebook into a production service, use this Skill to produce a data contract, baseline scorer, promotion gate config, canary rollout plan, and observation ledger as reviewable artifacts. ## Quick Start Ask the agent to use the mle-workflow skill to plan a production pipeline for your model, starting with the prediction contract and data contract.
