When does optimization not require experimentation? Seems like there would be a time..