ISO 22514-2 contains a table that crosses five methods for estimating location and five methods for estimating dispersion against eight time-dependent process models. Eighty combinations, of which thirty-eight are marked usable.
More than half the table is empty, and the absent content is precisely the value-adding content.
The estimator behind most Cpk figures in circulation (dispersion recovered from the average subgroup range through a d2 constant) is admissible for one model: constant location and dispersion, with a normally distributed instantaneous distribution. The standard designates this model A1 and states that a process conforming to it is under statistical control.
A tool wearing through a shift is a different model, as is a change of material lot, and as is a fixture progressively loosening.
None of this makes the arithmetic incorrect. It makes the arithmetic conditional, and that condition is rarely documented where it can easily be referenced when the time comes to review the result. Clause 7 states the requirement directly: where process performance or capability statistics are used for process qualification, the report shall state the calculation method used for location, the calculation method used for dispersion, the number of values on which the calculation rests, and the measurement uncertainty.
Of those four items, the capability reports I encounter in practice carry the number of values.
The 2026 edition is the third, replacing the 2017 edition after technical revision, and it added three process models to the classification along with a recommendation naming the one pair of estimators admissible under every model in the table. A capability procedure written against the second edition is therefore working from a smaller set of models and predates that recommendation.