Question 310
At fictional Northshore Public Library, a Snowflake data and AI engineering team is conducting an AI retrieval and source-citation assurance review under an approved model lifecycle and release-control process. The separately composed technical incident is: An insurance team extracts a numeric answer from long policy documents, but its system saves only the number and loses the passage and document version that justified it. 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: B (UNVERIFIED)
Preliminary individual option explanations — NOT expert validated
Correct Answer
Persist source metadata, test absent-answer cases and require human signoff for consequential extracted figures. Extracted answers need provenance and contextual validation, particularly for numeric or policy-sensitive facts.
Incorrect Answers
This response concerns Custom inference model deployment parity, which is different from the stated failure. An insurance team extracts a numeric answer from long policy documents, but its system saves only the number and loses the passage and document version that justified it.
Sentiment interpretation with irony is not the cause of the reported failure: An insurance team extracts a numeric answer from long policy documents, but its system saves only the number and loses the passage and document version that justified it.
Token budget exceeded by prompt assembly alone does not establish the needed evidence: Reviewers can trace sampled extracted numbers to exact effective policy passages and identify cases where the value is absent or qualified.