Question 170
A production Snowflake Gen AI workflow must handle the following: A legal operations bot is generating plausible but nonexistent paragraph references. What check would reveal the relevant functional or governance failure rather than a cosmetic success?
Explanation
Correct Answer
“Claim-level attribution accuracy, source link validity and abstention when no supporting passage exists” is supported by this requirement: A legal operations bot is generating plausible but nonexistent paragraph references. RAG trust depends on verifiable grounding, not decorative citations. Verify it using claim-level attribution accuracy, source link validity and abstention when no supporting passage exists.
Incorrect Answers
“Grammar score even if facts conflict with the source” would not meet the requirement that a legal operations bot is generating plausible but nonexistent paragraph references. RAG trust depends on verifiable grounding, not decorative citations. The more reliable approach is to require answers to reference retrieved source identifiers and validate that each substantive claim is supported by its cited passage.
“Presence of citation-shaped text regardless of linked content” would not meet the requirement that a legal operations bot is generating plausible but nonexistent paragraph references. RAG trust depends on verifiable grounding, not decorative citations. The more reliable approach is to require answers to reference retrieved source identifiers and validate that each substantive claim is supported by its cited passage.
“Model confidence wording without retrieved evidence” would not meet the requirement that a legal operations bot is generating plausible but nonexistent paragraph references. RAG trust depends on verifiable grounding, not decorative citations. The more reliable approach is to require answers to reference retrieved source identifiers and validate that each substantive claim is supported by its cited passage.