A statistical preprint proposes classification testing, an error-controlled alternative to binary significance tests that can leave results inconclusive.