What “good” looks like in expert review

“Incorrect. The model missed the key issue and the answer is not good enough.”

Everything in that is true and none of it is useful. It does not say which issue, why it matters, whether the error is one of knowledge or reasoning, or what a correct answer would have contained. A research team reading a hundred of these learns nothing they can act on.

“The conclusion is defensible but the reasoning is not. It identifies the correct treatment pathway, but arrives there by ruling out the alternative on the basis of the patient's age alone — which is not a sufficient discriminator here and would produce the wrong answer in a patient ten years younger. The response also omits any safety-netting advice, which the brief lists as required. A correct answer would identify the two competing diagnoses, distinguish them on the investigation findings rather than demographics, and state what should prompt re-presentation.”

Same verdict. Vastly more value. It separates conclusion from reasoning, names the specific flaw, explains when the flaw would bite, and states what a good answer contains.

Is the conclusion correct? Is the reasoning that produced it sound? Would acting on this be safe or professionally defensible?

An answer can pass the first and fail the second and third. Marking it correct because the endpoint matched is the most common calibration error, and the one first-submission feedback most often addresses.

A precise three-sentence review beats a rambling page. What matters is that a person who is not a specialist in your field can read your explanation and understand exactly what went wrong and why it matters.

If your review could have been written by someone without your qualification, it probably has not used it.