A county GenAIOps board requires a canary week when updating foundation models. What deployment pattern satisfies that policy?

Dual-running old and new foundation-model deployments during the canary week satisfies a county GenAIOps board policy that requires a canary week when updating foundation models. Side-by-side deployments keep the prior FM available while a limited or parallel path exercises the new version, enabling comparison of quality, safety, latency, and incident rate before full cutover. That pattern gives the board evidence and a fast rollback path if the new model misbehaves in real civic traffic.

Promoting the new foundation-model deployment to production before the canary week fails because it completes the production cutover before the required canary observation, leaving no dual-run week for comparison and rollback. Testing the new foundation-model deployment in staging for the canary week fails because staging-only exercise is not a production canary in real civic traffic. Scoring mirrored traffic using the new foundation-model deployment during the canary week fails because shadow scoring alone does not dual-run old and new deployments for production serving, comparison, and rollback.

Exam caveat: canary/dual-run is more specific than staging-only testing or scoring mirrored traffic. When policy explicitly requires a canary week, choose dual-run of old and new FM deployments rather than promoting to production before the canary.

Operational check: deploy the new FM alongside the old, shift a controlled traffic fraction or shadow comparisons during the canary week, review board metrics, then complete cutover or roll back while both deployments still exist.