問題文
A workflow must extract structured data from customer emails and use it to update a system of record. Which design keeps this reliable?
選択肢
- Store the raw emails and let a person enter the values into the system of record, since a human can resolve the ambiguous cases that no schema would cover
- Extract into a defined schema, validate the values in code, and update the system of record only for those values that have actually passed the validation
- Summarize each email and store the summary in a free-text note on the record, so that nothing from the customer's wording is lost when the values are interpreted
- Have the model update the system directly through a tool with write access, since the model already knows the field names and can correct an obvious mistake as it writes