It revealed a regression - return code not the same as v2.x for invalid PDF/A. It's also not easy to get the return code out of ruffus. Will need to tweak the final step of the pipeline.
