Recently, I saw that people were tagging their input records (test records in git repos) specifically so that later data loaders would reject those records in appropriate conditions. I forget what the tech was called but it was interesting.