The visitSummary files currently persist the detector for each row, but it turns out that this makes the files 140x larger than otherwise, and take 50x longer to read than otherwise. Persisting the detector in here is definitely not necessary; all we need is the id and that is persisted as the row index.