CGI models, ghosts and gorging on data – dissecting PR measurement with the PRCA