Question 90
At fictional Meadowford Care Network, a Snowflake data and AI engineering team is conducting an accountable generative AI operations change with legal, privacy and information-security review for any restricted inputs. The separately composed technical incident is: An enterprise plans to change embedding models for existing knowledge vectors, but its developers intend to mix old and new embedding outputs in the same similarity index without compatibility tests. 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
Version vector schemas, enforce model and dimension constraints, and rebenchmark retrieval before every embedding upgrade. Embedding vectors encode a model-specific representation; identical dimensions alone do not guarantee comparable semantic geometry.
Incorrect Answers
This response concerns Structured analytics versus document search routing, which is different from the stated failure. An enterprise plans to change embedding models for existing knowledge vectors, but its developers intend to mix old and new embedding outputs in the same similarity index without compatibility tests.
Prompt injection in retrieved business content is not the cause of the reported failure: An enterprise plans to change embedding models for existing knowledge vectors, but its developers intend to mix old and new embedding outputs in the same similarity index without compatibility tests.
Custom model artifact governance alone does not establish the needed evidence: Vector-shape tests and recall measurements confirm that all compared vectors use compatible representations and new indexing restores acceptable retrieval quality.