The useful engineering question is therefore not “Did 3.6 Flash win?” It is “Which differences are sufficiently supported to justify a production test?”