Unless I'm missing something, it's not the task of counting rows - it's claiming that their process catches the majority of data that should make it into a record in the final dataset, and produces few duplicates.