You're in spec, but... (part 3)

...you're running on luck.

Customer drawing comes in: bore diameter 50.00 ± 0.10 mm. You pull the last twenty parts off the lathe to see how you're doing. Every one passes. Closest call was 50.08 — still inside.

You quote the job and run it. Two weeks later you're sorting parts at the customer's dock.

Part 1 was a shift. Part 2 was a drift. This one's neither — the process doesn't change at all. The problem was there from the start.

The chart looks fine

Here's the I-chart for those twenty parts. The spec limits are in purple, control limits in red.

Process is "in control" — no points outside the control limits, no obvious patterns. Every measurement is between the spec lines too. If you're scoring this on "did parts pass," it seems fine.

But look at the gap between the data and the spec lines. Observation 9 sits at 50.08 — 0.02 mm from the upper spec. Observation 7 hits 49.94, 0.04 mm from the lower. The process isn't comfortably inside the tolerance — it's filling most of it.

The control chart can tell you the process is stable. It can't tell you whether stable is good enough.

And notice that the control limits are outside the spec limits. This means that normal process variation is wider than the tolerance allows.

The histogram shows what's happening

The histogram gives a different view of the same data. The bars are the parts you actually measured. The green curve is the shape of the process — it shows what measurements you are likely to see.

The bars all sit between the purple spec lines. You measured twenty parts and twenty parts passed.

The curve, however, doesn't sit completely between the spec lines. That's the process telling you what the next thousand parts look like — some of them won't fit between those lines.

You got twenty parts in spec because twenty parts isn't very many and you got a little lucky.

What the numbers say

Cpk — a common capability number — is 0.57 on this run. That predicts about 7% out of spec, roughly 1 in 15 parts. A usual minimum is 1.33.

So even though the process is stable and predictable, the process itself is too wide for the spec. The fix is upstream: a better machine, a tighter fixture, a cooler shop, less material variation. Something that narrows the green curve until it fits comfortably inside the purple lines.

Predictable isn't the same as acceptable

A stable process does what it always does. Whether what it always does is good enough for your customer is a separate question — and it's the one capability analysis answers.

Try it yourself — the not capable example on zonesix.ai/learn shows the same data with the full Cpk breakdown.