Chapter 2 of 6 · Architecture

Model governance, data residency, quota, and lifecycle

Governed foundation 5.5 hours in a non-production POC

Chapter 2 of 6

Architecture at a glance#

One external approval maps to one Azure child deployment. deployment-profiles.json carries the settings needed by preflight and Bicep. Preflight checks the exact Foundry scope, operator role, model availability and lifecycle, Responsible AI policy, processing location, quota, and resource-scoped what-if. A mismatch stops the run.

An approved model choice moves through preflight, deployment, review, and a keep, replace, or retire decision.

The deployment profile pins the model version with NoAutoUpgrade. The governed-agent deployment consumes the approved deployment name and model coordinates.

The approval register records the approved publisher, asset ID, source, hosting route, review decision, and processing-location boundary. The policy parameter file reads its allowed values, so the policy assignment and approval decision stay aligned.

Design choices and tradeoffs#

DecisionChosen approachCost or limit
Approval recordKeep the full review in the decision system and deployment inputs in GitThe approval ID must match both systems
Service factsRead lifecycle, availability, quota, and the named Responsible AI policy during preflightMissing CLI fields require a named manual check
Version changesPin exact coordinates with NoAutoUpgradeThe owner must start replacement before retirement
Deployment policyAssign the built-in approved-model and eligibility policies from the approval registerAudit needs review before Deny blocks a deployment
Control reachGovern this version-controlled deployment pathOther authorized paths remain open

Architecture guidance#

Session 03

Model governance, data residency, quota, and lifecycle slide deck