Skip to main content
ML-MODEL-MONITORING5 MIN READ

Canary a retrained model

Use prechecks, slice gates, and canary monitoring before promoting a retrained model.

Release pressure The candidate ranking model wins offline by 0.7%. The campaign starts in six hours, and the old model is still serving. Promotion is a risk decision; the offline metric is only one input. Canary discipline Precheck -> Slice gate -> Canary -> Promote Use the candidate model as a controlled release. The question is not whether it won offline; it is whether it meets quality, slice, and service guardrails under live traffic. Shortcut Ship to 100% because the leaderboard improved. Promotion criteria are written before the team sees live results. A canary is only useful when rollback criteria…

Read the full lesson

Sign up free — one personalized lesson every day, matched to your role and goals.

Already have an account? Sign in

← Back to library
Contact us