China High Precision Machining Sourcing: What to Verify Before You Pay

A ±0.005 mm tolerance is a statement made at 20 °C, on a calibrated CMM, in a temperature-controlled room at a factory you cannot see from your desk. When you source high precision machining from China, that statement crosses an ocean before the parts do. Twelve hours of flight, a change of humidity, a different gauge, a different person holding it, and the only thing you still control is the order in which you release money and the evidence you demand before each payment moves. This guide is written for that position. It breaks a precision claim into five evidence layers and maps each layer to a payment milestone you can actually enforce. It also shows you which China-specific checks — third-party inspection that reports to your inbox, live CMM measurement you watch, retained samples you re-measure yourself — close the gap between a factory's number and a number you can trust.

Why is the risk in a China precision order a measurement risk, not a price risk?

Price risk is the easy one. You compare three quotes, you understand why a ±0.005 mm feature costs more than a ±0.05 mm feature, and you buy from whoever priced the job honestly. Price risk resolves itself on paper. Measurement risk does not. A tight tolerance is not a property you can hold in your hand; it is a claim about a measurement result, produced by a specific machine, a specific instrument, and a specific room at a specific moment. Every one of those conditions is on the other side of the payment, and most of them are invisible to you until a part lands on your bench.

That inversion is what makes cross-border precision buying different from buying the same part domestically. At home, if the dimension is wrong, you pick up the phone and a person drives over with a gauge. Sourcing high precision machining from China removes the short loop: the wrongness travels in the same container as the parts, and by the time you can prove it, the payment you should have gated has already left. So the question "what do I verify before I pay" is really a question about which measurement claim each payment is buying, and whether you can re-verify that claim with evidence that does not pass through the seller's hands.

This article takes the layered view. Rather than a flat checklist of certificates and promises, it treats a precision order as a stack of evidence layers, each one weaker or stronger than the last, and it suggests how to let each payment milestone follow the evidence, not the calendar.

What does a ±0.005 mm band actually promise, and where is that promise made?

Before you can verify a claim, you have to know what the number contains. A callout of ±0.005 mm means the feature must land inside a window ten micrometers wide — 0.005 mm on either side of nominal. Ten micrometers is roughly a tenth of the thickness of a human hair, and it is a number that only exists relative to three things: a reference temperature, an instrument with a known calibration, and a defined measurement point.

The reference temperature is the least glamorous and most consequential part. International practice in dimensional metrology fixes the standard reference temperature at 20 °C — the value behind ISO 1 — because metal moves with temperature. Aluminum 6061-T6, the most common precision-machining alloy, expands at roughly 23.6 µm/m·K (the figure the site's aluminum-versus-stainless decision guide carries from ASM reference data). Steel expands at roughly half that rate. Neither is zero, which is why a credible CMM report states the temperature at which the measurement was taken, and why a serious metrology room holds the part and the machine near 20 °C before the probe touches it.

The instrument and the measurement point matter just as much. A ±0.005 mm callout on a bore needs a gauge whose own uncertainty is a small fraction of the ten-micrometer window; a caliper that resolves to 0.01 mm cannot judge it, while a CMM in a controlled environment can. And the measurement point matters because a feature is not a single number: a shaft measured at one axial position may read differently 5 mm away if the form is imperfect. The guide to which tolerances and finishes you can actually verify on arrival walks through exactly this boundary feature by feature.

Hold onto that picture. Every sentence a Chinese precision supplier writes about "holding tight tolerances" is a shorthand for this bundle: a temperature, an instrument, a calibration, a measurement plan. The high precision machining overview on this site unpacks that shorthand in numbers before any supplier enters the picture. Your job before paying is to pull the bundle apart and ask which pieces of it you can check — now, from a desk, and again after the parts land.

Which five layers of evidence sit between a supplier's word and a number you can trust?

Every claim in a China precision machining order arrives in one of five evidence layers. They run from the weakest — words — to the strongest — batch data you can re-audit yourself. The practical skill is knowing which layer you are looking at. Buyers who overpay mistake a weak layer for a strong one; buyers who get burned treat a strong layer as if it were weak.

Layer	What it contains	What it proves	Where it falls short
L1 — Capability statements	Verbal and email claims: "we hold ±0.005 mm," "our quality is excellent," promises made in sales conversations	Nothing you can act on; at most a signal of how the supplier talks	Costs nothing to produce; identical from a real shop and a shell company; carries zero payment weight
L2 — Certificates and machine lists	ISO 9001 / IATF 16949 (product design excluded) certificates, CMM calibration certificates, equipment tables with model numbers	The facility has systems and equipment — the certificate numbers can be checked against the issuing body's register	Proves "the shop has a CMM," not "the shop measured your batch"; certificates describe systems, not outcomes; a scanned certificate is easy to reproduce, so the number is the part that matters
L3 — Process documents for this order	First-article inspection (FAI) report, CMM data from the actual first piece, SPC records from the run	The batch was made under a documented process and was measured before it left	Self-reported: the same factory that made the parts also wrote the report; numbers can be smoothed or cherry-picked, and you have no independent witness
L4 — Physical evidence you can re-measure	Retained first-article samples, traceable batch numbers, parts or samples shipped ahead, third-party laboratory reports issued to you directly	A thing exists that you or your nominated laboratory can measure with your own instruments, on your own schedule	Proves the sample, not necessarily every piece in the batch; costs a little money and a few days to run
L5 — Batch-level data you can audit	Per-piece or per-batch measurement records with actual values, operator and timestamp, linked to batch numbers — the raw material behind SPC	The claim generalizes across the whole run, not just the first piece; you can sample the records and check them against the parts	Only as strong as the linkage: if the records do not carry batch numbers, they attach to nothing

Read the table as a ladder. L1 and L2 tell you a supplier could do the work. L3 tells you they say they did the work. L4 is where the evidence stops passing through the seller's hands — a retained sample, a third-party report addressed to you, a live measurement you watched. L5 is where the evidence becomes a population you can audit instead of a single flattering data point. A serious precision shop produces all five without being asked twice, because its automotive and aerospace customers already require them. The quality-control pages for this facility show what a documented chain — CMM verification, SPC at Cpk ≥ 1.67 on key characteristics — looks like when it is real.

How should your payment milestones line up with the evidence layers?

The payment structure is where this framework stops being theory. The principle is simple: release money only when the evidence layer you agreed to has actually arrived, and never let a calendar date substitute for it. What follows is a structural suggestion — a starting point to negotiate, not a promise of any supplier's terms and not a guarantee of any outcome. Treat the percentages as a frame, and the triggers as the enforceable part.

Milestone	Evidence gate	Share (structural suggestion)
Order and material commitment	L1–L2 desk checks pass: entity identity matches the contract party, certificate numbers verify on the registrar's register, CMM calibration is current, machine list matches the process your drawing needs	Roughly 30% — funds material and booked machine time; nothing more
First article and batch run	L3 arrives: FAI report against the drawing revision, CMM values for the critical dimensions, SPC evidence from the actual run	Roughly 40% — released against process documents for the order, not against a promise of them
Shipment and re-verification	L4–L5 close the order: retained first article shipped with the batch or a third-party report issued to your inbox, batch numbers linking records to parts, your own re-measurement or a laboratory's confirming the critical dimensions	Roughly 30% balance — held until the numbers you can check actually check

Three comments keep this honest. First, the deposit exists to buy material and schedule, and material evidence — a mill certificate for the alloy, linked to the batch — is the artifact that should accompany it. Second, the middle tranche is the one most buyers release too early: releasing it "when production starts" instead of "when the FAI and SPC arrive" gives the factory your money before the process documents exist. Third, the final balance is the only handle you still hold over the evidence layers L4 and L5, which is exactly why the retained sample and the third-party report belong before the last payment, not after it. A supplier that documents its work has no reason to resist this shape; a supplier that needs the full amount up front is asking you to fund the risk it is not willing to share.

Which China-specific checks let you verify a precision claim without flying to Dongguan?

Third-party inspection is mature in China, and it is the strongest tool a remote buyer has — when it is set up correctly. The detail that decides whether it protects you is the reporting line. If the factory pays for the inspection and the report lands in the factory's inbox, the factory is your inspector's client, and the chain of custody runs through the seller. The version that cuts the loop is simple: you appoint the inspection company — an international firm such as SGS or TÜV with a local China operation, or a laboratory in your own country — and you pay for it. The report is sent directly to your email address; it never passes through the factory. Ask for the measured values on the critical dimensions, not a summary page, and you have a document that did not pass through the interested party's hands at any point.

Live measurement is the cheap bridge between L3 and L4. A factory that will measure the first article on its CMM while you watch over a video call is offering you the next best thing to standing in the metrology room. The conditions matter more than the call itself: the part number should be visible, the probe should touch the features your drawing calls critical, and the nominal-versus-measured values should update on screen in real time. Ask that the temperature display be visible, and ask to record the session. A single watched measurement does not prove the whole batch, but it does prove that the first article existed, was measured, and was measured the way the report later claims.

Retained samples shipped to you are the verification that needs no trust at all. The strongest version of this on a precision order is to have the factory keep the first article and one piece from each batch — physically marked, batch-numbered. Ship those samples with the order, or ahead of it, so you can re-measure them at your own bench or send them to a laboratory of your choice. The question that separates a real evidence chain from a theatrical one is disarmingly simple: "what number did you measure?" A factory that ships its first article to you knows that question is coming and answers it in advance. One that charges extra for the sample, or finds reasons it cannot be spared, has just told you what it thinks of independent verification.

Two tools round out the set. XRF verification at receiving — a handheld gun that reads the heavier alloying elements — confirms the bar that became your part is the grade the mill certificate claims, which closes the material leg of the evidence chain on your side of the border. And because China is a twelve-plus-hour flight from most buyers, a recorded, timestamped video of the measurement environment becomes a document you can revisit months later. Film the CMM room, the temperature readout, and the part in the fixture — detail a photo that arrives through a chat app never captures.

Why can the same part read differently at your dock than it did in the factory?

Here is the trap that catches buyers who do everything else right. Your parts arrive, you take them straight to the bench, and the micrometer says the critical dimension is out of band. Before you write the angry email, run the numbers on temperature, because a ±0.005 mm claim was measured at 20 °C, and your receiving dock is rarely at 20 °C on the day a shipment lands.

The arithmetic is not subtle. Aluminum 6061-T6 expands at about 23.6 µm/m·K. Take a 100 mm critical feature — a locating land, a bore centerline spacing, a datum face separation — and let it sit ten degrees away from 20 °C: the feature moves by roughly 0.024 mm, about twenty-four micrometers. The entire ±0.005 mm band is only ten micrometers wide. A ten-degree difference is more than twice the width of the tolerance, and it is not an exotic scenario — it is a container sitting in the sun at a port, a cargo hold at altitude, or a warm receiving area in July. Even two or three degrees of offset pushes a long aluminum feature several micrometers, a large share of the band, before you have measured anything at all.

The discipline that follows is not to distrust your micrometer but to distrust the moment you use it. A credible arrival check gives the parts time to reach the reference temperature: open the packaging, set the parts and your gauge in the same room, and let them soak until both read a stable temperature — hours, not minutes, for a solid block. Steel parts move less than aluminum for the same swing, but they move; the habit is identical. Measure the critical dimensions only after the soak, record the room temperature next to your readings, and state both when you compare numbers with the factory.

Temperature is the largest and least-credited cause of "their report says good, my bench says bad," but it is not the only one. Your gauge has its own calibration history; their CMM was calibrated by a laboratory you have not met. You may be measuring a different axial position on a shaft, or a different point on a bore, than the factory measured. The drawing revision on your desk may not be the revision they machined. The sequence that keeps disagreements productive is always the same: confirm revision, confirm the measurement point, confirm calibration, confirm temperature, and only then compare the numbers. Most "the factory is lying" moments dissolve into "we measured different things at different temperatures," which is a conversation, not a crisis.

Which red flags in a precision supplier's evidence chain should stop a payment?

Some patterns in the evidence chain should halt the process on the spot. Each one below comes with a check that costs about five minutes and needs nothing more than email access and a search bar.

• The payment account does not match the contract party. The invoice names one company; the bank details name another, or a personal account. Five-minute check: ask, in writing, for the account holder's legal name and the registration number behind it, and confirm they match the entity on the contract. Wire transfers are fast and nearly irreversible; the account name is your last gate.
• Certificates are offered as scans, but the numbers do not verify. A genuine ISO or IATF certificate carries a certificate number issued by a certification body that keeps a public register. Five-minute check: find the registrar named on the certificate, open its verification page, and type the number in. A number that returns nothing, or a certificate that names no registrar, is a decoration.
• Retained samples are refused, or priced as an extra. A factory that documents its work ships its first article as a matter of routine. Five-minute check: ask for the sample policy in writing, in the same email as the quote. A policy that cannot name a sample is a policy that does not want you measuring.
• The CMM report has a blank temperature field, or a temperature far from 20 °C. The temperature is the context that makes a ten-micrometer number meaningful. Five-minute check: ask for the room temperature at measurement and the calibration reference of the CMM, and ask whether the factory applies thermal compensation. A shop that cannot answer has not controlled the condition the number depends on.
• The quoted tolerance "relaxes" as the quantity grows. The price per part may legitimately fall with volume; the tolerance band does not, because capability is a property of the process, not of the order size. Five-minute check: ask for the process-capability value on the critical features at the quoted volume. A supplier that moves from ±0.005 mm to ±0.02 mm "because you ordered more" has just described a process that cannot hold the band at volume.
• One hundred percent prepayment is demanded. Material and schedule justify a deposit; nothing justifies the full amount before any evidence layer exists. Five-minute check: propose the milestone structure from the earlier section. The response — negotiation or refusal — tells you more than any certificate.

Now the honest counterweight, because a verification framework that assumes every supplier is a fraud is as useless as one that assumes none are. Most export-oriented Chinese precision shops are not running on forgery; the automotive supply chain alone — factories operating IATF 16949 (product design excluded) under third-party surveillance, shipping to customers who audit them — has made genuine documentation the norm for a large tier of the market. The framework in this article is cheap screening against the bad apples that do exist at the low end, and it doubles as a method for disputes with good suppliers, where the problem is usually measurement method, not intent. The same discipline applies to you: your own gauge can be out of calibration, your own measurement point can differ, and a buyer who accuses before checking his own method has skipped the step his supplier is being judged on. Verify first, on both sides, in that order.

Which clauses turn this evidence chain into terms your supplier can sign?

Everything above becomes enforceable in exactly one document: the purchase order. The clauses below are a skeleton you can adapt — plain English, no legal boilerplate, written so a factory's quality manager can read them without a lawyer. Have your own contract review adapt them to your jurisdiction; the point here is that the verification chain is negotiable only before the order is signed, not after the parts are in dispute.

• Batch measurement reports. "Each batch shall ship with a CMM dimensional report for the critical features, stating the measurement temperature, the calibration certificate reference of the CMM, the operator, and the date." This one clause forces L3 to carry its own context.
• Samples and third-party access. "The first-article part and one part from each batch shall be retained, batch-numbered, and made available for shipment to the buyer or to a laboratory the buyer names. The buyer may appoint a third-party inspection company for pre-shipment inspection; the supplier shall provide access and the report shall be issued to the buyer directly."
• Payment conditioned on re-verification. "The final payment releases after the buyer's re-measurement of the retained samples at 20 °C, or after a third-party report confirms the critical dimensions, within the acceptance window stated in this order."
• Dispute measurement. "In the event of disagreement between the supplier's measurements and the buyer's, the parties shall submit the contested features for measurement by a jointly agreed third-party laboratory, whose CMM results shall be binding for determining conformity."

None of these clauses signals distrust to a factory that already produces the evidence for its other customers. They simply make the verification chain a term of the deal instead of a favor you have to ask for later, and a supplier who signs them has told you, in writing, that its evidence will survive contact with your micrometer. That is the cheapest insurance a high precision machining order can carry.

Why should a legitimate precision supplier welcome a buyer who verifies?

The last question is the one that tells you whether you picked the right factory. A shop that runs real metrology treats verification the way a bank treats an audit: as a routine cost of being trusted. Its automotive customers audit its process; its CMMs are calibrated on a schedule by laboratories it can name; its first-article samples are artifacts it already keeps. A buyer who asks to see the evidence chain is, to that supplier, a normal professional doing normal due diligence. That is the same customer, in fact, who is least likely to become a dispute later, because the method was agreed before the parts were made.

A supplier that flinches at the questions in this article is not protecting a legitimate process from an unreasonable buyer. Flinching at naming its registrar, at shipping its first article, at a video call around the CMM, at a third party reporting to your inbox, that is protecting something from being seen. The layers in this guide do not exist to make honest suppliers prove themselves endlessly. They exist to make the difference between an honest supplier and a bad one visible in a week of emails — before the first wire, while the only thing at risk is a deposit you could have kept.

Frequently Asked Questions

Can I trust a Chinese factory's ISO certificate?

A genuine ISO 9001 or IATF 16949 (product design excluded) certificate tells you the factory runs a documented quality system — it does not tell you this batch was measured correctly. Verify the certificate number against the registrar's public register, confirm the scope covers machining, and check the surveillance dates are current. Then remember what layer you are looking at: a certificate is L2 evidence. It proves the shop has systems; the CMM report on your parts is what proves the outcome.

Can a factory tamper with a third-party inspection report?

Only if the report passes through the factory's hands. The setup that removes the risk is simple: you appoint the inspection company, you pay for it, and the report is emailed directly to you by the inspector — the factory never holds the document. Ask for measured values on the critical dimensions rather than a summary page, and if you want extra assurance, call the inspector's office to confirm the report is genuine before you release the balance.

Why does my re-measurement disagree with the factory's outbound report?

Work through the four causes in order before drawing conclusions. Confirm the drawing revision matches. Confirm you are measuring the same point on the same feature. Confirm both instruments are calibrated and traceable. And confirm temperature: their reading was taken near 20 °C, and yours only counts after the part has soaked to a stable temperature. Most disagreements dissolve into method differences; the ones that survive all four checks are the ones worth a third-party measurement.

Does it matter whether ±0.005 mm parts travel by air or by sea?

Not for the dimension itself. Neither airfreight nor ocean freight changes the size of a rigid metal part; both can damage parts through mishandling if packaging is wrong, and both expose the cargo to temperature swings that matter for measurement but not for the part's permanent dimensions. Choose the route on cost, time, and packaging risk — then treat the arrival check the same way either way: unbox, let the parts and your gauge soak to a stable temperature, and only then compare numbers with the factory's report.

Should I buy my own micrometer to verify precision parts?

Yes — a calibrated micrometer is the cheapest layer of independence you can own, and it is the right minimum instrument for checking a ±0.005 mm diameter at your bench. Know its limits: a micrometer reads a two-point diameter and says nothing about roundness, position, or true position, which is why the form and location callouts on your drawing still belong to a CMM report or a laboratory. Buy the micrometer, calibrate it on a schedule, and reserve the CMM for the features hand tools cannot see.

What should I look for if a factory offers to show the CMM measurement live?

Accept, and watch for the details that make it evidence: the part number and drawing revision on screen, the probe actually touching the critical features, the nominal-versus-measured values updating in real time, and the temperature display in view. Ask to record the session. A live measurement proves the first article was measured the way the report claims; it does not prove the rest of the batch, which is what the batch-level data and your own sample checks are for.

Is a Chinese factory's XRF report enough to prove the material grade?

No — XRF reads the heavier alloying elements, so it separates a 316 stainless from a 304 or a free-cutting steel from a plain carbon grade, but it cannot measure light elements such as carbon or sulfur. Use it as your independent spot check that the material is the grade the mill certificate claims, and keep the certificate itself — with its heat number — as the document that carries full grade identity and mechanical properties.

How do I dispute a measurement disagreement without destroying the relationship?

Treat it as a method problem before it becomes a trust problem. Send the factory your numbers, your instrument's calibration certificate, the room temperature at measurement, and the exact point you measured, and ask for the same detail on their side. Most disagreements resolve at this stage. If they do not, invoke the dispute clause from your purchase order — a jointly agreed third-party laboratory — and let an independent CMM decide. A supplier that cooperates through that sequence is showing you how it will behave for the life of the program.