A library GenAI index includes draft policies never approved for public guidance. What should the auditor emphasize?
Select an answer to reveal the explanation.
Short Explanation
Draft policies are backstage rehearsal — not the show residents should see. Keep the GenAI index to approved guidance only. Otherwise the bot may sound official while quoting unfinished rules.
Full Explanation
Restricting the library GenAI corpus to content with documented approval status suitable for public guidance works because generative indexes will surface whatever they are given with an authoritative tone. Draft policies, working papers, and unreleased rule changes are internal work product; if indexed, the chatbot can quote speculative requirements as if they were adopted guidance for residents. Source-governance controls that gate inclusion on approval status, effective dates, and public-use designation prevent the model from laundering incomplete policy into resident-facing answers. The correct emphasis is corpus hygiene tied to records of approval, not model cleverness after the fact.
Prioritizing draft policies in retrieval ranking using recency signals for resident queries fails because recency does not equal approval, and boosting recently edited drafts can push incomplete or superseded proposals ahead of adopted guidance for residents. Improving answer completeness by widening the chunking window for draft policy sections fails because extracting more text from unreleased drafts increases the risk of quoting unadopted language, and completeness built on unapproved content is not trustworthy public guidance. Surfacing draft policies with formatted citations to support transparency in new daily resident chatbot answers fails because a citation does not make an unapproved draft into authorized guidance; it only gives unreleased policy an undeserved air of official provenance.
Exam caveat: some drafts may be intentionally published for public comment; those need explicit “comment draft” labeling and retrieval instructions so the bot discloses status rather than silently treating them as final rules. Operational check: maintain an approval metadata field on every corpus document, exclude items lacking public-guidance status from the live index, and spot-check chatbot citations against the approved-policy register after each corpus refresh.