Testing theory and the minimal detectable bias