considered to be of equal quality, and this provides the simultaneous control of the closeness to the target and the variability around the mean. So when the mean drifts away from the target, then the standard deviation has to be lower in order to maintain the coverage.
Similarly to bioequivalence testing where inequivalence is the null hypothesis, we have defined null hypothesis as a batch quality out of specification. This associates the type I error with the practical most important error; namely, the undesirable event that a batch is released but is outside specification. This is yet not the usual approach within the CMC arena as it is in clinical sciences, but it is necessary to provide statistical rigor.
Since the quality of batches released to the consumer is of the greatest importance, it is appropriate to set the null hypothesis at out of specification because this then has to be refuted by data with high confidence in order for the batch to pass. And this is key to understanding our approach to the view, and I hope that Walter will touch upon this hypothesis framework a bit more, so it will be crystal clear at the end of the day.