There isn’t a way to select a desired schema version, if I understand you correctly, but there should be no need to do so. Pipelines don’t currently use Avro’s schema resolution mechanism to convert each input record to an instance of a “reader schema”. They instead directly attempt to find each requested path (as per the (col <- path::in::schema DEFAULT [val], ...) clause) in each new input schema which they encounter, loading the DEFAULT value for paths which can’t be found, or NULL if there’s no explicit DEFAULT.