A survey includes an open-ended question, and one reviewer alone assigns every response to a category. Which improvement most directly strengthens the reliability of the coded results?