做完實驗算出的「最佳組合」，通常一次都沒被做出來過。它是預測值，不是實測值——所以田口方法規定你必須再跑一次確認實驗。這一步是推論與事實的分界線，卻是中文教材裡最常被跳過的一段。寫了一篇直交表入門，用一個起司蛋糕從頭走到尾。The “optimal setting” you compute from a designed experiment has usually not been run. It’s a prediction, not a measurement. Taguchi’s method says: go run it. That …