Question 450
A Snowflake platform team receives the following requirement: A data scientist prototypes feature engineering in a Notebook but keeps critical model inputs only in transient local notebook variables. Which preventive control would most directly reduce recurrence?
Explanation
Correct Answer
This guardrail directly addresses snowflake notebook reproducibility: an interactive notebook state is not a durable governed pipeline output. The prevention is: Treat notebook work as reproducible data and code rather than a permanent live session. A separate subsystem check would not prevent the specific recurrence.
Incorrect Answers
Document content extraction needs a supported unstructured-data processing capability. “Validate document extraction accuracy and sensitive-file permissions before production use” would address that alternative cause, not the condition in this question: an interactive notebook state is not a durable governed pipeline output.
“Validate underlying data permissions and app ownership before sharing internal applications” responds to this separate finding: application sharing requires governed Snowflake object and role access, not account administration. This question instead needs “Treat notebook work as reproducible data and code rather than a permanent live session”, so the alternative does not meet the stated condition.
Literal equality filtering does not provide semantic retrieval over meaning-equivalent text. “Measure retrieval relevance, freshness and access controls when indexing business text” applies to that separate problem. The observed case is a data scientist prototypes feature engineering in a Notebook but keeps critical model inputs only in transient local notebook variables; choosing this alternative would address the wrong failure mode.