A tight tolerance on a drawing means little if the supplier cannot hold it on every shift, every setup, and every machine in the cell. Cpk machining data is how a capable shop proves that claim with numbers. Instead of showing one lucky inspection, capability indices summarize a history of real measurements and predict how comfortably the process sits inside the tolerance band. Regulated buyers ask for that evidence because parts that pass sampling today can still hide a drifting process tomorrow. This guide explains Cp and Cpk, the meaning of the common 1.33 and 1.67 thresholds, and what honest capability proof looks like from a CNC supplier.
What Does Cpk Machining Capability Measure?
Cpk, the process capability index, compares a measured characteristic to its specification limits using the process average and the standard deviation of real measurements. It asks a simple question: how many standard deviations separate the mean from the nearest tolerance limit? Dividing that distance by three gives Cpk. A Cpk of 1.0 means the natural spread of the process just touches the closest spec limit. Below 1.0, some output is expected outside tolerance. Above 1.33, the process keeps comfortable clearance on both sides.
The index turns a pile of inspection data into one comparable figure. When a buyer reviews a Cpk machining report for a bore diameter, they see centering and spread for that feature, on that machine, with that program, during the study window. The math applies to any measurable characteristic, from a positional tolerance to a flatness value, as long as the data is real and the process was stable while it was collected.
Cp and Cpk Tell Different Stories
Cp looks only at the width of the process spread compared with the tolerance width. A process can have a healthy Cp and still produce scrap if its average sits near one limit. Cpk fixes that blind spot by penalizing an off-center mean. The relationship is strict: Cpk never exceeds Cp, and the gap between the two figures is a centering problem, not a variation problem.
Engineers use that distinction to pick the corrective action. When Cp is fine and Cpk is low, the shop adjusts the offset, checks the workstop, or updates wear compensation. When Cp itself is low, the fix is structural: sharper toolpaths, fresher inserts, better workholding, or a cooler, more stable environment. A Cpk machining summary that shows both indices is far more useful than either number alone. Choosing which characteristics deserve that treatment starts with the print, and our GD&T basics for CNC buyers article explains how datums and modifiers shape the measurement plan.
Why 1.33 and 1.67 Show Up on Purchase Orders
Many automotive-derived and aerospace-adjacent requirements call for a minimum Cpk of 1.33 in ongoing production and 1.67 for new processes until capability is demonstrated. The numbers are not arbitrary. Under normal-distribution assumptions, a centered process at Cpk 1.33 predicts an escape rate measured in tens of parts per million, while 1.67 pushes the estimate below one part per million. Both targets describe defect rates that sampling inspection alone cannot reliably guarantee.
The two-tier approach has a behavioral purpose as well. A new process carries unknown bias from setup and tool break-in, so buyers demand extra margin while evidence accumulates, then settle at 1.33 once the process has proven stability. Capability figures layered on top of a drifting process, however, are meaningless. Demonstrating stability during the study window belongs to statistical monitoring, covered in our overview of CNC machining quality control.
How a Shop Demonstrates Genuine Capability
Credible capability work follows a sequence. First, the shop selects key characteristics from the drawing, usually the features that control fit, function, or safety. Next it verifies the measurement system: gauge repeatability and reproducibility, or GR&R, shows that the gauge adds little error relative to the tolerance. Data collected with a weak gauge distorts every index calculated from it. Then comes the sample plan. A common convention borrowed from automotive PPAP practice calls for 25 subgroups of five consecutive parts from a qualified process, roughly 125 readings.
Only then does the study move to calculation, and critically, to a stability review. The readings go onto control charts that confirm the process behaved predictably while the data was gathered. A Cpk machining figure computed from an unstable process is a number without a referent, since the index assumes predictable behavior. The measurement plan itself depends on how the tolerance is drawn; our CNC machining tolerances guide covers how tolerance choices affect which readings matter.
Where Buyers Actually Use Capability Numbers
Capability data earns its place in specific sourcing situations. High-volume orders use it to justify sampling plans instead of 100 percent inspection. Regulated buyers fold Cpk machining studies into supplier approval: medical device purchasers expect statistical evidence under ISO 13485 practices, a workflow we outline in our medical CNC machining article, and automotive programs write thresholds directly into PPAP requirements. In aerospace work, capability evidence supports the process validation that AS9100 expects.
Scorecards are the second use. A key characteristic whose index slips from 1.8 to 1.3 across quarterly studies is an early warning that tooling, gauging, or setup discipline has changed. That trend conversation is more productive than arguing about one outlier reading, because it focuses attention on the process rather than on a single part.
Reading Capability Numbers With Healthy Skepticism
Every summary statistic can be flattered. Some characteristics do not follow a normal distribution at all. Surface roughness, for example, is bounded at zero and often skewed, so an index computed as if it were normal will mislead; our guide to surface roughness Ra explains how to read that measurement family properly. Small samples overstate confidence, cherry-picked dimensions inflate results, and a single study captures one moment rather than a durable condition.
The buyer defense is simple: ask for the raw data and the control chart alongside the index, require a stated sample plan, and insist on re-verification after any meaningful process change. A shop that shows its work invites exactly that review. A supplier that quotes Cpk machining results without context is asking you to take process capability on faith.
What is a good Cpk value for CNC machined parts?
Most customers ask for 1.33 as the floor in stable production and 1.67 for new or recently changed processes. Values well above 1.67 on a machined feature are strong but not always economic, because pushing variation lower usually means more frequent tool changes, faster gauging, or tighter environmental control. The right target depends on what the characteristic does and what failure would cost.
What is the difference between Cpk and Ppk?
Cpk uses the variation within subgroups to estimate short-term capability, while Ppk uses the total spread of all readings, including long-term shifts between subgroups. In practice Ppk is the more conservative statement of what the process actually delivered over the study period. A common convention asks for both, with the long-term figure held to a higher target during initial studies before the process settles at its ongoing floor.
Which dimensions need a capability study?
Key characteristics and anything that has failed inspection before are the natural candidates, and measuring everything is wasteful. Choose the features that control fit, sealing, rotation, or safety, then confirm the shortlist with the supplier quality engineer. The datum scheme on the drawing usually reveals which dimensions carry the functional load.
