They reference parquet files, not sure if it's only CSV or CSV even figures in that heavily other than the first iteration before migrating to spark