What actually drifts
It is almost never the material. It is the setup.
The programme was rewritten. Small shops frequently reprogramme from the drawing on a repeat order rather than recalling the proven programme. New toolpaths, new feeds and speeds, new results.
The fixturing changed. A first article held in a soft-jaw setup and a production batch held in a vice will not produce the same flatness or the same distortion.
The machine changed. Proven on a well maintained mill, repeated on whatever had capacity. Spindle condition and rigidity vary between machines in the same shop.
Tooling changed. A different insert grade or a worn tool at the end of a batch produces chatter, poor surface finish and features that wander off nominal.
Inspection is end-of-batch only. If the first part is checked and the fiftieth is checked and nothing between, you get an average, not control. Tool wear is a ramp, and the parts in the middle are where the ramp crosses the tolerance.
The operator changed. Undocumented knowledge is the most fragile thing in a machine shop. Real process control means the setup does not depend on who is standing at the machine.
Surface finish and cosmetics are where the arguments happen
Dimensional failures are objective. Cosmetic failures are where buyer and supplier fall out, because "acceptable" was never defined.
Surface finish must be a number on the drawing, not an adjective. Ra 0.8 is a specification. "Good machined finish" is a future dispute.
Say where it applies. A finish call-out that applies to the whole part when you only need it on one sealing face inflates the price and multiplies the chance of a reject.
Agree how it is verified. Comparator block, profilometer, or by agreement to a signed boundary sample. Say which, before the order.
If appearance matters, send a boundary sample. A physical reference for "worst acceptable" ends more arguments than a paragraph of description ever will.
Three ways to control repeat quality, compared
| Approach | What it costs | What it gives you | Best for |
|---|---|---|---|
| Trust and inspect on goods-in | Your inspection time on every delivery | Late detection. You find the problem after paying for freight | Low-value non-critical parts |
| Locked process with in-process checks | Some setup discipline at the supplier, a slightly higher price | Consistency, and traceability back to a cause when it does drift | Repeat production parts and anything that goes into an assembly |
| Full PPAP or first article approval to AS9102 | Real money and several weeks up front | Formal control of every characteristic and every change | Aerospace, medical, automotive, safety-critical |
For most SME buyers the middle row is the right answer, and it is the one most rarely asked for.
How to lock a part down (six things to ask for)
- Programme retention. Ask the supplier to confirm in writing that the proven programme is retained and reused on repeats, and that reprogramming requires notification.
- Fixture retention. Ask who owns the fixtures and where they are stored. Put ownership in the purchase order terms.
- In-process inspection frequency. First off, then a stated interval, then last off. State the frequency rather than leaving it to the shop.
- A change notification clause. Any change of machine, tooling, material source or sub-supplier gets notified before it happens. This single clause prevents most drift.
- Certificate of conformity naming the drawing revision. The exact revision, not just the part number.
- A boundary sample for anything cosmetic.
None of that is exotic. All of it is cheaper than a rejected batch. The paperwork side, certificates and revision control, has its own guide.
What this looks like in practice
Every part TrueNorth ships is inspected before dispatch rather than after delivery. That is a deliberate cost, and it is where drift gets caught.
On a square flange order the parts came back from the machinist with chipping on a machined face, a surface finish that failed the drawing requirement of Ra 0.8, and evidence of tool wander. All three are classic drift signatures, and all three point at tooling condition and setup rigidity rather than at the operator.
The disposition was scrap and replace at our cost. The supplier received a formal corrective action request with two working days to contain and ten working days to return a full 8D root cause. "Operator error" is not an accepted root cause on that form, because it is a description of the symptom.
The customer's line kept running. That is the only measure that counted that week. Being told early matters too, and that discipline is covered in the supplier communication guide.
We also hold our suppliers to an 18-month warranty on defects in the purchase terms, which is longer than the 12 months we offer downstream. The gap is deliberate.