A water utility compares lab-narrative embeddings and must avoid distance skew from unnormalized vectors. What should they apply before comparisons?
Select an answer to reveal the explanation.
Short Explanation
VECTOR_NORMALIZE is lining up the rulers before you measure—normalize lab-narrative vectors so distance math stays fair.
Full Explanation
A water utility comparing lab-narrative embeddings must apply VECTOR_NORMALIZE before distance comparisons when unnormalized vectors would skew distance by magnitude. DP-800 includes VECTOR_NORMALIZE so municipal pipelines can place embeddings on a comparable scale before ranking or thresholding distances. Normalization prevents longer-magnitude vectors from dominating comparisons when the intended signal is direction/similarity of meaning in lab narratives. ROUND on unrelated numeric lab readings only affects scalar measurements and does not normalize embedding vectors. FORMATMESSAGE formats strings for messaging and performs no vector math. SOUNDEX on the utility name is a phonetic text function and is not a stand-in for vector normalization. Exam caveat: some embedding models already emit normalized vectors; blindly double-normalizing is usually harmless for unit length but teams should confirm the model’s contract and the distance metric in use so operational checks match training assumptions. Operational check: take a pair of lab-narrative embeddings, compare distances before and after VECTOR_NORMALIZE, confirm magnitude skew is removed for the chosen metric, and add normalization as an explicit step in the utility’s comparison stored procedure or view.