A SaaS program manager runs a support-triage agent that classifies tickets and drafts replies. After a prompt edit caused wrong routing last week, leadership requires traceability of what ran, quick rollback, and documented changes. Which versioning practice should the manager require to enable traceable rollback of the agentic workflow?

The correct practice works because Git stores the full workflow definition as versioned code where semantic tags mark releasable states and changelog entries record rationale for every prompt change, so the manager can map a ticket run to an exact commit, redeploy a prior tag within minutes after a routing regression, and show leadership a documented audit trail. The cloud snapshot practice fails the documentation concept because timestamped copies support fast restoration but provide no semantic meaning or reason for change, leaving teams unable to differentiate major routing logic edits from minor wording tweaks during review. The log versions with manual rebuild practice fails the rapid recovery concept because run ID linkage plus a changelog delivers traceability and documentation but forces engineers to reconstruct the prior workflow by hand, which is slow and error prone and does not meet the quick rollback mandate. The approved changelog with tagged files practice fails the runtime traceability concept because it preserves intent and restorable artifacts yet never binds execution records to the specific version that ran, so the cause of last week's misrouting cannot be proven. As an exam caveat, do not mistake observability or approval records alone for versioned rollback since traceable rollback requires a runnable version, a run-to-version link, and change documentation together. As a practical check, pick a recent ticket, resolve its run ID to a commit hash and tag, redeploy the previous tag in staging, and confirm the changelog explains the prompt difference.