Anna Severin explains how her team used machine learning to try to assess the quality of thousands of reviewers’ reports.