Production readiness

Before production, confirm all of the following:

• The workspace, data sources, and keys are isolated from experimentation.
• The current version passed representative evaluation cases.
• Tool scopes, spend limits, concurrency, and approval boundaries are explicit.
• Each external write is idempotent or has a documented remediation path.
• Run traces, alerts, and on-call ownership are in place.
• You know which version to roll back to and how to stop new work.

For the first production week, read successful traces as well as failures. A result that looks correct for the wrong reason is the earliest signal to fix.