What Cpk value is actually acceptable?

A press operator brings a capability study to the quality desk and asks the question every shop asks eventually. It came out at 1.28, is that a pass? It is a fair question with an unsatisfying answer, because 1.28 on its own is neither a pass nor a fail. Whether it clears depends on what the customer asked for, whether the number is short term or long term, and whether the process was even holding still when the readings were taken. The rule-of-thumb thresholds everyone repeats are a starting point, not the verdict.

Where 1.33 and 1.67 came from

The familiar numbers are not arbitrary. A Cpk of 1.33 corresponds to four standard deviations between the process average and the nearer specification limit, which the automotive world settled on decades ago as a reasonable floor for an ongoing process. A Cpk of 1.67, five standard deviations, is the tighter bar often demanded for safety-related or otherwise critical characteristics, and for new processes at approval where you want headroom before the process has had a chance to drift. Below 1.0 the process is producing outside tolerance and everyone agrees it fails. Between 1.0 and 1.33 sits the uncomfortable middle where the parts might be in spec today but there is no margin for the normal wandering every real process does. So the thresholds are sound guidance. They are just guidance about a generic process, and you are not running a generic process.

A Cpk on an unstable process is a number about nothing

Here is the trap the threshold conversation walks people into. A team chases 1.33, hits it on paper, and signs off, without ever asking whether the process was in control while the study ran. Capability assumes stability. The whole calculation rests on the process behaving like one consistent process rather than several different ones taking turns. If the average was shifting or the spread was changing during the study, the Cpk you computed describes a process that does not really exist, an average of conditions that were never all true at once. It can read 1.5 and still ship bad parts next week, because the number never captured the thing that was actually moving. Confirm the process is stable first, on a control chart, and only then does the capability index describe something real enough to hold against a target.

Make the target and the reading meet in one place

A capability target is only useful if the reading it is judged against carries its own context, the characteristic, the specification limits, the part revision and enough samples over a fair window. When capability lives in a spreadsheet, that context leaks and the number drifts free of the thing it was supposed to describe. If you want to check where your own readings land, the free Cp, Cpk, Pp and Ppk calculator gives you all four with a plain-language read on each. In VoraControl, operators capture against the released control plan characteristic, so the specification limits and revision travel with every reading, and capability is trended straight off the capture stream against those limits rather than recomputed from a study nobody can reproduce. The target still comes from your customer. What changes is that the number you compare to it is one you can stand behind.

So the honest answer to what Cpk is acceptable is two sentences long. It is whatever your customer specified for that characteristic, at the stage you are reporting, and no rule of thumb overrides a stated requirement. And it only counts once the process was stable enough for a single capability number to mean anything at all.