A business requirement for an internal HR assistant reads: the assistant should give trustworthy answers about leave policy. The solution architecture already places the approved policy text in a document store the assistant can search, and a named policy owner is available to adjudicate answers. Which restatement is a usable functional requirement?
- AThe assistant should be tuned and re-prompted until the reviewers on the project agree that its leave answers feel trustworthy during day-to-day use of the internal tool.
- BThe assistant must be built on the largest model available to the team, on the grounds that answer quality on policy questions rises with model capability.
- CFor each question in a fixed review set, the assistant must return an answer whose cited policy clause matches the clause the policy owner selects for that question. Correct
- DThe assistant must carry a system-prompt instruction telling it to answer only from approved policy documents and to decline any question those documents do not cover.
Why A is wrong: Tempting because it names a review step and sounds rigorous, but it restates the vague word in the business requirement rather than replacing it, so there is no condition an engineer can build against or a test can fail.
Why B is wrong: Tempting because capability does correlate with quality, but this fixes an implementation choice instead of stating the behaviour required, and it can be satisfied in full while the assistant still answers from stale or unapproved text.
Why C is correct: Correct: it converts the business intent into an observable pass or fail condition, using a fixed input set and an authoritative human judgement, so the requirement can be tested repeatedly and regressions are visible.
Why D is wrong: Tempting because the wording sounds like a control, but an instruction describes intended behaviour rather than a checkable outcome, so nothing in it tells a tester whether a given answer met the requirement.