It sounds like this is something special that happens only with this data set since you say you have not had this happen before. I suggest sending the example or something similar to technical support so they can investigate further.
Thanks Tonya. With 77 views at time of writing and not even a hint of possible solutions, that's enough to know this is a rare occurrence. I have since figured out that the more data points I exclude from the calculation of the spec limits, the less offset I observe. This is a workaround though and doesn't explain what's actually happening. I will raise with technical support as you suggest. Many thanks.