The next question should be: “Do these predicted y ̂ and s ̂ values at the so called optimal settings of A_(+1) C_(-1) produce results that will satisfy the customer?” If the answer is no, we might consider setting A larger than +1 coded and/or C less than -1 coded (this is called extrapolating). The cheapest approach would be to reduce C because a larger A would increase cycle time and reduce our throughput. Once the decision on C and A is made, we should predict the ideal y ̂ and s ̂ and then return to the process of confirming our predictions. The casiest way to check for confirmation is to build the prediction interval y ̂ ± 3s ̂ which should contain all confirmation response values. If the confirmation runs produce a response value clearly outside the prediction interval, we say that we have “failed to confirm” our models.