A product operations lead oversees an agentic workflow that drafts weekly release notes from merged pull requests and closed issues for a mobile app with strict marketing claims review. Drafts sometimes invent features or omit fixes. Which supervision practice ensures that agentic workflows meet requirements for accuracy before publishing?

The correct practice succeeds because it enforces accountable human review before any publication, grounds every claim in the authoritative evidence for that release, and applies defined acceptance criteria to satisfy strict marketing claims review consistently; verifying drafts directly against linked pull requests and closed issues specifically prevents invented features and catches omitted fixes, while the pre-publish gate ensures errors never reach customers. The automated validation distractor fails on oversight type because semantic similarity checks can flag obvious drift at scale but cannot interpret nuanced claims, approve regulated language, or take accountability for accuracy. The quarterly audit distractor fails on control timing because reviewing samples long after release may improve future prompts yet permits inaccurate weekly notes to publish in the interim, missing the before-publishing mandate. The roadmap-based approval distractor fails on evidentiary grounding because roadmaps and briefs reflect planned intent and positioning rather than merged work, so review would preserve messaging tone while leaving hallucinations and omissions undetected. Exam caveat: when a stem highlights fabrication or omission, prioritize verification against linked source artifacts over generic quality, speed, or style improvements. Practical check: block publishing until an approver attests each bullet links to a specific pull request or closed issue and each in-scope fix appears in the notes.