Sample question 1 — Prompt Engineering & Structured Output
A research pipeline pulls title, authors, abstract and sample size from life-science papers into required JSON properties. Reviewers found rows where a number was invented because the paper never stated it and the contract gave the model nowhere to admit the gap. The four properties must stay present for downstream tables while a missing value becomes visible. Which change fixes the defect?
- A. Make each extraction field nullable so an uncertain value is reported as null with a short reason.
- B. Give the model a separate confidence number per field and keep every value non-null in the contract.
- C. Mark every property optional so the model can drop any field it cannot read with full confidence.
- D. Let the model append uncertainty notes to a free-text explanation field while required fields stay filled.
Answer: A. Option A is correct: make each extraction field nullable so an uncertain value is reported as null with a short reason. Keeping the key present while allowing an explicit null separates 'the paper does not say' from 'the paper says zero', which is the distinction the reviewers could not see. Option B still forces a number into every row and only adds a second opinion beside it. Option C removes the key, so a table cannot tell a gap from a dropped column. Option D hides the admission in prose no table reads.
