Question 210
At fictional Ashford Health Analytics, a Snowflake data and AI engineering team is conducting an AI retrieval and source-citation assurance review using approved synthetic testing data instead of confidential customer records. The separately composed technical incident is: A research knowledge app drops relevant passages when similarity scores fall below a handpicked fixed threshold, but it has not measured how different query types score. 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
Track relevance across query segments and reassess thresholds after embedding models, text domains or retrieval settings change. Vector similarity is a ranking signal, not a universally calibrated probability of source truth or answer quality.
Incorrect Answers
This response concerns Keyword and semantic retrieval complementarity, which is different from the stated failure. A research knowledge app drops relevant passages when similarity scores fall below a handpicked fixed threshold, but it has not measured how different query types score.
Reranker prioritizes outdated policy version is not the cause of the reported failure: A research knowledge app drops relevant passages when similarity scores fall below a handpicked fixed threshold, but it has not measured how different query types score.
Summarization loses contractual exceptions alone does not establish the needed evidence: Measured retrieval precision and recall improve on rare and common query sets, and low-evidence queries produce reviewable abstention outcomes.