This allows us to normalize bad data, and is the first step towards splitting note types into separate tables.