A school-district GenAIOps pipeline must stop production when Foundry prompts no longer match source control. What enforcement fits?

Failing the build when production prompts diverge from the Git main branch enforces prompt integrity for a school-district GenAIOps pipeline. The pipeline compares the live Foundry (or deployment package) prompt hash to the approved main-branch artifact and stops promotion or flags drift when they differ. That prevents a staging check against a feature branch from letting silent edits into the lasting tutoring-bot personality reach production without returning to source control parents and curriculum owners can review.

Blocking the release if staging prompts differ from the Git feature branch checks the wrong environment and branch—it validates staging against unstable work instead of stopping production when prompts diverge from main. Failing the build if Foundry models diverge from the Git main branch polices the wrong artifact—it tracks model versions instead of prompt text and does not stop production when prompts diverge from source control. Blocking promotion if production prompts differ from the Blob audit copy uses the wrong source of truth—it compares production to a downstream backup instead of the Git main branch and does not enforce source-control match.

Exam caveat: Build failure on drift assumes a reliable extraction of the production prompt and a clear canonical branch. Emergency break-glass edits may be allowed with time-boxed exceptions, but the pipeline must still require back-port and re-approval—not permanent bypass.

Operational check: Wire a pre-deploy or continuous drift job that fetches production prompt text, hashes it, and diffs against main; break the build or page on-call on mismatch; log who last changed Foundry configuration; and periodically restore production from Git to prove recovery works for the tutoring bot.