Scientific progress depends on results that can be repeated – yet across disciplines, too many experiments fail that test.