A "good" Cpk result doesn't mean there won't be nonconforming parts. It's supposed to help predict the number of nonconforming parts produced by a stable, normally-distributed process. As we know, there are serious problems with using Cpk for predictions (or using it at all).
We also do not have enough data to determine if the characteristic exhibits a stable, normally-distributed process. If there is random springback issues, then using the total variation equation to determine what contributing variations are affecting the population may be necessary to figure out the true capability. Dumping data into an equation is never enough.
So, we do not have a lot of evidence to support the use of Cpk as a viable estimate of capability, which can also be part of the problem.
We don't know at this point if what the customer found is what the customer asked for in terms of Cpk.
But, I was trying to give some basis to the issue of Cpk versus bad parts in the population. I also agree that it is not a precise measure of capability by any means.
Last edited: