If your real question is “Did Group A do better?” say that in advance. Direction locks the “tail” of your test. Declare “greater,” “less,” or “different” before the data tempts you. Software will gladly hand you any p-value you ask for, including the wrong one. Good science is a pre-commitment device: define the claim, then test it. When direction matters (faster load time, higher conversion, fewer errors), choose one-sided; when any change matters, choose two-sided. Clarity beats cleverness. Aim first. Shoot second.
