A team set out to validate software used in fMRI analysis, but ended up invalidating one of neuroscience's most common testing procedures.