Question 240
At fictional Granite Public Works, a Snowflake data and AI engineering team is conducting a trusted self-service analytics redesign using only first-party public technical references and independently written facts. The separately composed technical incident is: An AI assistant retrieves multiple policies and reorders them for semantic relevance, but the final top passage is an archived superseded policy with obsolete requirements. Assume only these documented facts and use safe, authorized production-change procedures; exact current Snowflake feature syntax and region or account availability require current review. Which repeatable assurance or control is most relevant to preventing this situation?
Explanation
Writer-proposed answer: D (UNVERIFIED)
Preliminary individual option explanations — NOT expert validated
Correct Answer
Include validity metadata, document lifecycle tests and source version audits in retrieval regression checks. Reranking optimizes relevance among candidates and does not itself establish that a document is authoritative or current.
Incorrect Answers
This response concerns Answer extraction without source attribution, which is different from the stated failure. An AI assistant retrieves multiple policies and reorders them for semantic relevance, but the final top passage is an archived superseded policy with obsolete requirements.
Embedding dimension mismatch in a vector column is insufficient to demonstrate this result: Known change-history questions select the active approved policy, while retired documents are excluded or clearly labeled historical.
Structured completion schema validation is not the cause of the reported failure: An AI assistant retrieves multiple policies and reorders them for semantic relevance, but the final top passage is an archived superseded policy with obsolete requirements.