Question 60
At fictional Willowhaven Municipal Office, a Snowflake data and AI engineering team is conducting a source-system modernization and AI search review under documented decision rights and minimum-privilege security requirements. The separately composed technical incident is: A research organization uploads several similarly named custom models to a registry, but no one can tell which training inputs, evaluation run or model version supported the deployed inference endpoint. 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
Require registry ownership, immutable version identifiers, signed evaluation gates and routine endpoint-to-model version reconciliation. Model registry governance provides an accountable relationship between training artifacts, evaluated versions and serving decisions.
Incorrect Answers
This response concerns Cross-region model inference consent and residency, which is different from the stated failure. A research organization uploads several similarly named custom models to a registry, but no one can tell which training inputs, evaluation run or model version supported the deployed inference endpoint.
Gen AI evaluation set representativeness is insufficient to demonstrate this result: A deployment audit traces the serving endpoint to an immutable approved model version, training evidence and evaluation record, and an authorized rollback reproduces the prior response.
Grounded employee policy assistant architecture is not the cause of the reported failure: A research organization uploads several similarly named custom models to a registry, but no one can tell which training inputs, evaluation run or model version supported the deployed inference endpoint.