Design a Promotion Gating Policy Across Model Risk Tiers

Your company has three model types in the registry: an internal "suggested article" recommender (low stakes, retrained nightly), a churn-prediction model that drives which customers get a retention discount (medium stakes, financial impact but no regulatory exposure), and a credit-limit model (high stakes, regulated, direct financial impact on individual customers).

1. Propose a different staging → production promotion policy for each of the three models, and justify the differences.
2. The credit-limit model's quality gate shows a challenger version scoring notably better than the production champion — well outside the range of normal retraining variation. Under your proposed policy, what happens next, and why should a large improvement trigger scrutiny rather than fast-tracked promotion?
3. A well-intentioned engineer proposes applying the credit-limit model's strict human-review policy to all three models "to be safe." What's the downside of doing that?