The machine passed incoming inspection. The parts didn't.
Six weeks after the riggers left, the aluminum housings started drifting. Nothing dramatic — a couple of thousandths on the bore diameter, then a few more on the second op. Not enough to fail first article. Plenty enough to fail the third shipment.
The machine was new. The laser report said ±0.005 mm. The purchase order said "ISO 230-2 compliant." Everything matched on paper. The parts didn't match the paper.
That's the version of this problem most buyers haven't been told about. It isn't that you picked the wrong model, or that the price was too high, or that the freight forwarder dropped it. It's that "the machine meets spec" and "the machine makes good parts in your building, on your fixturing, with your operators, in July" are two different claims that quietly get treated as one.
I handle quality and brand compliance for an industrial components supplier. Roughly 30 to 40 machine evaluations a year cross my desk, plus spec review on every capital purchase above $50,000. Over four years of doing this, we reject about one in five first deliveries on capital equipment — not because the machines are broken, but because the documentation doesn't actually support the claim being made. That's a strange sentence to write, and it took me a while to believe it myself.
The number on the spec sheet has conditions attached (nobody lists them)
"±0.005 mm positioning accuracy" isn't a property of the machine. It's the output of a test, run under a specific protocol, on a specific day, at a specific temperature. ISO 230-2 defines how you measure positioning accuracy and repeatability — where to put the target, how many cycles, how to handle approach direction, whether to compensate for thermal drift. What it doesn't do is force anyone to tell you which of those conditions they used.
So two machines with identical numbers on the brochure can behave completely differently. One was measured at 20°C in a climate-controlled room, warm, with thermal compensation active, over a 400 mm segment of a 1,000 mm travel. The other was measured cold, over the full travel, without compensation, and rounded down from a worse figure. Both statements are technically true. Only one tells you what you're buying.
This matters more by machine type than most sourcing conversations admit. On a vertical machining center, the positioning spec mostly governs Z-depth control and interpolated contour accuracy — annoying to get wrong, usually recoverable with tool compensation. On a horizontal machining center, the number that actually governs your parts is often pallet changer repeatability, which doesn't appear in the positioning figure at all. And on a basic 3-axis CNC machining center running simple prismatic parts, positioning accuracy is almost beside the point — squareness and spindle thermal growth will eat you alive long before the linear spec does.
Five-axis is where it gets expensive. Simultaneous 5-axis accuracy isn't captured by any single linear figure. You need rotary axis accuracy, and then you need to know how the two rotaries behave together at the pivot point — a different test entirely, usually ballbar or volumetric, and almost never included in a standard quote package. When someone quotes a 5-axis machine on positioning accuracy alone, they've answered a question you didn't ask.
The incentive problem underneath CNC machining center sourcing
Here's the part that took me longer to accept than it should have. Sourcing agents and dealer networks aren't structured to care whether your machine holds tolerance in month eight. Their commission clears when the machine lands. Warranty, when there is one, typically runs 12 months on parts and labor — and the failure mode we're talking about, a machine that "passes" but never quite makes consistent parts, rarely trips a warranty claim. It isn't broken. It's just not as good as advertised.
That's not corruption. It's a timeline mismatch. The agent is measured on placement. You're measured on output. Those two things diverge somewhere around month four.
So when you're comparing CNC machining center manufacturers, the question that actually protects you isn't "who has the best reputation." It's "who will sign a contract that puts real money behind the acceptance criteria, and who will still answer the phone in month nine?"
I've made the other mistake myself. In 2023 we took delivery of a 3-axis machine and I skipped the re-test after it sat in an unheated bay for two days while installation dragged on. My reasoning was embarrassingly simple: it's the same machine, the numbers aren't going to move. They moved. Not much — but enough that the first production run of a ±0.02 mm bore came in at 0.031. We caught it on the third article, but the first two shipped. That was about $9,400 in rework and an uncomfortable phone call (the call was worse than the rework, honestly). I knew I should have re-run the test. I thought, what are the odds? The odds are pretty good when you change the thermal environment.
The third time a machine arrived with a clean factory report and then failed our own cut test, I finally wrote an incoming acceptance protocol. That should have happened after the first time. It's about four pages long now and it's the single most useful document I own.
What it actually costs
Let me put real numbers on this, because "quality problems are expensive" doesn't move anyone's budget meeting.
Scrap and rework. A machine that holds 80% of nominal tolerance instead of the 30% you planned for doesn't fail loudly. It drifts into the margins of your process capability, and your scrap rate goes from 1.5% to 6%. On a part with $40 of material and 20 minutes of cycle time, that's not a rounding error — it's the difference between a job that pays for the machine and one that doesn't.
Downtime while you argue. The worst stretch sits between "we think the machine is the problem" and "the manufacturer agrees." That window usually runs 6 to 14 weeks, and during it you're either running parts you don't trust or not running at all. If the machine is on a customer's program with a delivery commitment, you're burning goodwill at a rate you can't invoice.
The workaround tax. A drifting machine forces you to compensate — tighter inspection, more frequent offsets, slower feeds. You're now paying for the machine twice: once on the invoice, once in the productivity you gave up to work around it.
Resale value. If you ever sell it, the buyer's inspection will find what yours didn't. That discount comes out of your pocket.
For scale on the capital side: publicly listed dealer and auction data through 2024–2025 puts a new 3-axis vertical machining center roughly in the $45,000–$180,000 range depending on travels, spindle, and control; a horizontal machining center with a pallet changer typically starts near $250,000 and climbs past $600,000; a simultaneous 5-axis machine often lands between $300,000 and $1.2 million. Those ranges move — verify against current quotes before you budget. But the point stands: a $200,000 machine that underperforms by 30% is a $60,000 mistake you discover slowly, in scrap and overtime, over 18 months.
The fix is boringly contractual (which is why nobody does it)
I'm not going to walk you through a 40-step sourcing methodology. The fix is smaller than that, and mostly involves things you already know you should be doing.
1. Define the acceptance test, not just the spec. "Meets ISO 230-2" is not an acceptance criterion. "Passes ISO 230-2 positioning accuracy and repeatability testing at 20°C ±2°C, over full travel, after a 2-hour warm-up, on our fixture, with raw data provided" is. That difference is roughly one paragraph of contract language, and it changes the entire negotiation.
2. Run your own cut test before final payment. A machine can pass every metrology test and still chatter on your part geometry. Cut a standardized test piece — NAS 979 circle-diamond-square, or a part from your own production — after the machine is installed and thermally settled. Hold 15–20% of the purchase price against that test passing.
3. Ask for raw data, not conclusions. Any reputable manufacturer runs laser interferometry and ballbar testing. Ask for the raw files. A summary saying "passed" is a marketing document. Raw data with dates, temperatures, and axis-by-axis results is evidence. If they won't send it, that tells you something too.
4. Match the test to the machine type. Vertical needs spindle thermal and Z-depth verification. Horizontal needs pallet changer repeatability verified separately. Five-axis needs rotary accuracy plus volumetric testing at the pivot. One acceptance template across all three is exactly how you end up with a machine that technically passed and practically doesn't perform.
5. Keep a named contact past the warranty. Not a general support line — a person, with a name and an email and a track record of replying. It's the cheapest insurance in the whole deal and most buyers never ask for it.
None of that is glamorous. It's just the difference between buying a machine and buying the output the machine was supposed to produce. An informed buyer asks harder questions earlier, and the sourcing conversation actually gets shorter — not because you got a better price, but because you stopped discovering problems in month six.