Question 200
At fictional Moorland Cultural Institute, a Snowflake data and AI engineering team is conducting a governed document-grounding pilot while preserving enterprise data boundaries and a reviewable audit trail. The separately composed technical incident is: A policy document is split into fixed chunks that separate requirements from their exceptions, causing the retrieval system to supply only the restriction and omit the following qualification. 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
Recompute retrieval tests after any parser or chunking change and monitor source completeness and citation usefulness. Chunking shapes the evidence available to a generator; incomplete context cannot be repaired reliably by better prompting alone.
Incorrect Answers
This response concerns Sensitive translation with regional constraints, which is different from the stated failure. A policy document is split into fixed chunks that separate requirements from their exceptions, causing the retrieval system to supply only the restriction and omit the following qualification.
Token budget exceeded by prompt assembly is insufficient to demonstrate this result: Retrieval evaluations return the rule and its exception together for labeled test questions, with the final answer citing both passages.
Structured completion schema validation is not the cause of the reported failure: A policy document is split into fixed chunks that separate requirements from their exceptions, causing the retrieval system to supply only the restriction and omit the following qualification.