Cp or Pp, which capability index should you report?

A customer engineer opens your PPAP package, finds the capability study, and emails back inside the hour. This says Ppk 1.52, we asked for Cpk, please resend. It reads like pedantry until you notice the two numbers came off the very same measurements and still disagree, sometimes by a wide margin. Cp, Cpk, Pp and Ppk are not four labels for one idea. They answer two different questions, and reporting the one you were not asked for is a fast way to look like you do not know your own process.

Two questions hide behind four letters

Line the four up and the pattern is simpler than the alphabet soup suggests. Cp and Cpk use a within-subgroup estimate of variation, the scatter you see inside each small sample as it is taken, worked out from the average range or average standard deviation of the subgroups. Pp and Ppk use the overall standard deviation of every reading in the study from start to finish. That one difference in how the spread is measured is the whole story. Cp and Pp then compare that spread to the tolerance width and assume the process is centred, while Cpk and Ppk drop the centred assumption and account for how far the average has drifted toward the nearer limit, which is why the k versions are almost always the honest pair to quote. So the real fork is not Cpk against Ppk. It is within-subgroup against overall, potential against performance.

The gap between the two is the part worth reading

Most people pick one index, report it, and throw the other away. The more useful move is to keep both and read the distance between them. If Cpk and Ppk sit close together, the process behaves minute to minute much as it does across the whole run, which is another way of saying it is stable. If Cpk sits comfortably above Ppk, the process is capable in any given moment but wandering between moments, so something is moving the average around, a tool warming up, a material lot changing, an operator nudging a setting. That gap is telling you the trouble is not capability at all. It is stability, and tightening the short-term spread does nothing for a process that will not hold still. Chase the drift first, and the number you finally report starts to mean something.

A number is only as good as the context beneath it

None of the four indices is worth anything if you cannot say which characteristic it belongs to, which specification limits it was measured against, and which part revision was in force when the readings were taken. That context is what turns a number into evidence, and it is exactly what falls off when capability lives in a spreadsheet. If you want to work the maths on your own readings first, the free Cp, Cpk, Pp and Ppk calculator runs all four side by side and puts the verdict in plain words. In VoraControl, operators capture against a released control plan characteristic, so every reading already carries its part, revision, operation and specification limits, and capability is computed straight off that capture stream instead of a study filed months ago and hoped to still apply. The index becomes a reading of the work, not a document ageing beside it.