Audit answer-bank citation freshness against live Confluence docs

Keeps your RAG answer bank honest. Each stored answer is linked to a Confluence source page; this workflow periodically re-reads those pages, compares the current content against what the answer claims, and uses an LLM to decide whether the citation is still valid. Entries backed by changed or missing docs are marked stale so retrieval can skip them.

When support docs evolve faster than your knowledge base and you risk serving confidently wrong answers. Run nightly to guarantee every retrievable answer still points at a live, consistent source.

1. 1A nightly schedule starts the audit.
2. 2The workflow pages through all active entries in the Postgres answer bank.
3. 3For each entry, it fetches the cited Confluence page by ID and reads its current version and last-updated date.
4. 4A logic step compares stored vs. current version hashes to find pages that changed or 404'd.
5. 5For changed pages, an LLM judges whether the answer still matches the source.
6. 6Entries that fail verification are flagged `stale` in Postgres and a freshness summary is posted to a Slack review channel.