Model Staging and Promotion
You'll learn: How to safely manage a model's lifecycle from development to production through structured approval workflows and staging environments.
What Is Model Staging?
Model staging is the practice of moving models through defined lifecycle stages as they mature and gain approval. Think of it like a software release pipeline: you don't deploy code directly from your laptop to production—you test it in staging environments first. Models deserve the same discipline.
The typical stages are:
- Development/None: Model is being trained and experimented with
- Staging: Model passed initial tests and is ready for validation in a production-like environment
- Production: Model is actively serving predictions
- Archived: Model is retired but kept for auditing or rollback
Why Staging Matters
Without staging, you risk deploying undertested models that could produce incorrect predictions, degrade user experience, or violate compliance requirements. Staging creates a safety gate where stakeholders can validate performance on production-like data before committing.
Approval Workflows and Governance
Model promotion between stages should require explicit approval. In MLflow Model Registry (from your previous lesson), you'd transition a model version from "Staging" to "Production" only after:
- Validating metrics meet thresholds
- Running A/B tests or shadow deployments
- Getting sign-off from ML engineers, data scientists, or business stakeholders
- Documenting the decision and rationale
This creates an audit trail showing who approved what, when, and why—critical for regulated industries and debugging production issues.
Example Pattern
A data scientist registers a new model version → automatically tagged as "None" → after validation, promotes to "Staging" → monitoring team validates in staging environment → team lead approves promotion to "Production" → old production model moves to "Archived."
Key Takeaway: Model staging and promotion workflows enforce governance by requiring models to pass through validation stages with explicit approvals before reaching production, reducing risk and maintaining accountability.