OpenAI Search score calculation

Hi everyone,

Similar topics were discussed at Semantic Search Doubt yet how \Search endpoint calculates the score of the results is not addressed quite well.

I need to know what is the logic behind it so that I can set a meaningful threshold for the results that \Search yields for my queries. Because sometimes results that have score above 200 can be unrelated, but scores around 20 can be quite meaningful.

Good result with score way below 200:
*Q: What was the product/service that you provided this client? *
Answer with highest score(28.312): Solution company provided to the product with transaction histories of 4236 accounts containing over thousands of transactions.

Poor result with score > 200:
Q: Does high tech describe your product?
Answer with highest score(265.596):
High level of risk: The team recognised the risk of working directly with this data.

Thanks in advance,