The final dataset needs to be converted into CDISC ARS. Probably this should just be done by mangling the data in an extra pipeline step. - There might be limitations to the data model..