Read this before you read the numbers. This article quotes published third-party figures to show how a denominator changes an answer. We cannot audit those figures and they are not our data. Do not lift a rate, a threshold or a standard number from any article, including this one, into a purchase order. Ask the inspection body and the authority that will sign on your project.
Ask ten window suppliers for their defect rate and you will get ten numbers. Very few of them will be invented, and almost none of them will be answering the question you actually asked.
A defect rate is not a property of a factory, a country or a product family. It is the output of a measurement protocol. Change what counts as a defect, who counts it, what you divide by, or the day you stop counting, and the same shipment will produce a different number, often by an order of magnitude.
So this article does the unglamorous thing. It shows three genuinely published figures that all answer different questions, sets out the five inputs that produce any rate, explains why the sampling standard most buyers name does not measure what they think it measures, and ends with the clauses that make a rate measurable on your project.
Before you ask a supplier for a rate, it is worth seeing what rates already exist in public. Here are three, each real, each widely circulated, and each counting something else.
| The figure | Where it was counted | What it actually answers |
|---|---|---|
| A market report on coloured aluminium sliding windows cites a flight inspection in which only 58.4 percent of a random national sample of 1,200 finished batches met the full national standard in every respect | Random market sampling of finished product, all price tiers, whole country | How much of what is on the market complies. It does not describe what any named supplier will ship you |
| A Chinese industry quality white paper analysing 200 window and door projects reports an overall pass rate of 76 percent, with material quality highest and fabrication and installation lower; across more than 5,000 consumer complaints, hardware failure, opening problems, profile damage and leakage accounted for 62 percent | Installed projects, counted after installation and after occupation | Where projects go wrong. Note that two of the four leading complaint categories sit in installed work or in use, not in delivered product |
| A factory publishes that statistical process control keeps defects below 0.5 percent | Internal line output, at the factory gate, counted by the factory | What the factory saw when it looked. Not independently auditable, and not measured at your site |
Now look at the direction those three numbers run in. The market-wide figure is the worst, the installed-project figure is in the middle, and the factory figure is the best. That ordering is not deception. It is when the counting stopped.
Every gate a product passes moves defects from "not yet counted" to "counted". Count at the machine and you get the smallest number. Count at the container and it grows. Count at site receipt and it grows again. Count after the first winter rain and it grows again. The rate is a function of how long you keep counting, which is why a number quoted without a gate is not a number, it is a mood.
If two parties disagree about a defect rate, they are almost always disagreeing about one of these five things rather than about the windows.
| Input | What varies | If you leave it open |
|---|---|---|
| 1. What counts | A coating scratch, a colour step between batches, a sash that drags, a sealed unit that fogs. Each has to be classified critical, major or minor against a written list | The inspector counts one thing and the buyer reads another. This is the single largest source of inspection disputes |
| 2. Who counts | Factory final check, the buyer's agency, the installer at handover, the occupant after move-in | Four counters produce four numbers and every one of them is correct. The same split applies to factory swing door systems: the figure generated at the gate and the figure generated at handover are both real, and they are not the same figure |
| 3. What you divide by | Per unit, per sash, per lineal metre of frame, per square metre of glass, per opening, per project | A two-panel sliding window and a six-panel sliding door are not one denominator, and averaging them hides both |
| 4. When the count stops | Factory gate, pre-shipment, site receipt, completion, first rain, end of the defects period | The party who chooses the closing date chooses the number. Nobody says this out loud, and it decides everything |
| 5. What happens to what you found | Rework, sort, ｒｅｐｌａｃｅ, accept with a written concession | If the remedy is to sort before loading, the defect was removed and never enters any published rate. A factory that sorts hard can honestly report a gate figure near zero |
One practical consequence of input three: an order that mixes a two-panel sliding window with a 4 panel entry door is not one lot and should not have one number. Different panel counts carry different numbers of corners, rollers, interlocks and drainage paths, so their failure opportunities are not distributed the same way.
Most buyers write "inspect at AQL 2.5" into a purchase order and believe they have specified quality. Public guidance on acceptance sampling is unusually blunt about why that does not work.
An AQL of 2.5 does not mean 2.5 percent of your shipment may be defective. In ISO 2859-1 (the current edition was published in 2026; the United States equivalent is ANSI/ASQ Z1.4) it stands for Acceptance Quality Limit: the worst process average that will still usually be tolerated across a continuing series of lots before the sampling scheme starts rejecting them. The older phrase "Acceptable Quality Level" is still everywhere and still points at the same tables, but the word "limit" is the honest one. A limit is where you stop accepting, not a target you aim at.
Take a published worked example for normal single sampling: a lot of 5,000 units at General Inspection Level II gives sample-size code letter L and a sample of 200 units. At AQL 2.5 for the major class, the acceptance number is 10 and the rejection number is 11. Ten or fewer qualifying nonconformities and the lot stays inside the acceptance number; at eleven, the rejection number has been reached. Nothing in that arithmetic multiplies 5,000 by 2.5 percent. If you calculated 125 allowed defects, you calculated a number no table contains.
Sample sizes at General Level II, from the same public guidance: 50 units for lots of 281 to 500, 80 for 501 to 1,200, 125 for 1,201 to 3,200, 200 for 3,201 to 10,000 and 315 for 10,001 to 35,000.
Here is the part that should change how you read an inspection report. For the 200-piece plan above, the same public source states that a lot which is truly 2.5 percent defective is accepted about 99 percent of the time, and a lot which is 5 percent defective still passes about 58 percent of the time. Moving to Inspection Level III raises the sample from 200 to 315 and drops that second figure to about 39 percent. Still not zero, and that is the point.
Acceptance sampling is a decision rule with a known error rate, agreed in advance so that a buyer in one country, a supplier in another and an inspector in a third all reach the same verdict from the same evidence. It is not a measurement of your shipment. A passed AQL inspection tells you a sample satisfied a rule on one day in one building. It does not tell you the defect rate of the goods, and it never did.
The standard does not tell you whether a scratch, a colour variation, a broken closure or a safety hazard is critical, major or minor. Public summaries give the usual shape: a critical defect makes the product unsafe or non-compliant, is normally set at AQL 0, and a single one fails the lot; a major defect stops the product doing its job or would cause a customer to return it, commonly AQL 2.5; a minor defect is a finish flaw a customer would tolerate, commonly AQL 4.0. Each unit is counted once, at its worst defect.
Which of those three your particular scratch falls into is a decision you make, in writing, before the inspector draws the sample. Inspection guidance is consistent on where AQL instructions actually fail, and none of the failures is arithmetic: the lot is never defined; the supplier lays the samples out in advance; variants, colours and configurations are not represented; the buyer and the inspector disagree about what counts as major; and nobody agreed in writing what happens after a failed result. A correct table lookup cannot rescue any of those.
Two more things worth knowing. Sampling schemes include switching rules between normal, tightened and reduced inspection precisely because one pass is not a verdict. And Inspection Level III is normally recommended for a first order with a new factory, for anything expensive to recover from, and after any lot has failed.
Every defect has a birthday. Naming it is most of the work of assigning responsibility, and it is the reason a single rate cannot describe a project.
| Born | Typical examples | Who can see it, and when |
|---|---|---|
| At the factory | Wrong wall thickness, wrong glass build, colour step between batches, corner joint not sealed, a roller that is flat-spotted out of the box, a sealed unit whose perimeter seal fails | Visible during production and at pre-shipment inspection |
| In transit | Coating scuffed where a stillage touched, track deformed by strap pressure, moisture staining under wrapping, glass broken but held in place by the packing until it is opened | Invisible to the pre-shipment inspection. Visible only at site receipt, and usually discovered during installation instead |
| On site | Cutting, drilling and glazing done on site, packing and shimming, perimeter sealant, corner treatment, weep holes painted or caulked over during redecoration, a track used as a mixing spot | Appears after handover, and is almost always reported as a product defect first |
| In use | Roller and pile wear, adjustment drift, salt and chlorine exposure, blocked drainage, cleaning with the wrong agent | Appears over time, faster on a window opened twice a day than on one in a holiday house. The difference is duty, not quality |
This is the same reason that commercial swing door installation belongs in a separate quality file from the door itself: different counter, different gate, different number. One file records what was delivered, the other records what was done with it, and merging the two is how a good product acquires a bad rate.
| Claim category | How it is read | What settles it |
|---|---|---|
| Material or profile | Compare the delivered profile with the agreed alloy, wall thickness, dimensions, surface finish and inspection record | Mill certificate and incoming inspection, not a photograph |
| Cutting, machining, glazing, hardware fitting | Review the processing record and the hardware installation | Whoever performed the operation owns the outcome |
| Sealing, drainage, corner treatment, alignment | Inspect the seals, joints, drainage openings and site conditions | Responsibility follows control of the activity, not ownership of the product |
| Transport | Compare packing and loading records against the condition recorded at receipt | The receipt record, made on the day, with the delivery note annotated before the driver leaves |
| Mixed or uncertain cause | Work through contract, approved drawings, samples, reports, photographs and time-stamped assembly records | The evidence chain. There is no shortcut, and a label is not a finding |
Sliding products generate a distinctive complaint set, and a large share of it is not made of defects at all. If you write your defect list without this table, your rate will be inflated by maintenance items and understated on the two that matter.
| Symptom | What it usually is | Birth | In the factory rate? |
|---|---|---|---|
| Water standing in the outer track | The design working. A sliding frame is a managed drainage system, not a sealed box: the outer seal reduces infiltration but is not expected to stop it under wind pressure, and what gets in is collected and leaves through the weep holes | By design | No. The failure is water past the inner upstand, not water in the outer track |
| Water overflowing to the room side | Blocked weep holes first, then missing sill slope, then a seal that has lost compression | Factory or site | Usually not. Check the weep holes before anything else; they are the most common cause and the most commonly mis-repaired |
| Weep holes sealed over | Paint or caulk covering the exterior opening during redecoration. Invisible from inside, and it can happen after the window has passed every inspection it will ever have | Site | Never. And the classic wrong repair, sealing the visible gap, closes the drainage path that was working |
| Stiff or jerky travel | Dirt and debris in the track, dried-out lubrication, worn or flat-spotted rollers, or a sash that is not level on its adjusters | Factory only if the roller arrives damaged; otherwise use | Only the flat-spotted roller. Everything else is maintenance |
| Lock will not engage | The sash has dropped on its rollers and the keeper no longer lines up with the latch | Use, or factory if delivered mis-set | The reported fault is the lock; the actual fault is the roller height. Counting it as a lock defect produces a misleading rate |
| Fogging between the panes | Failed perimeter seal on the insulated glass unit | Factory | Yes. This is the one complaint that is nearly always a genuine product defect. The frame is usually sound and the sealed unit is replaced rather than the window |
| Condensation on the room-side glass | Warm humid indoor air meeting a cold surface | Use | No. This is the exact opposite claim to the row above, and the two are reported as the same thing constantly |
| Rattle in wind, or daylight at a closed sash | Pile or compression seals flattened, or adjustment drifted | Use, or factory if delivered out of adjustment | Maybe. Adjustment is a maintenance action, not a nonconformity |
| Pitting, or hardware corrosion | Environment: coastal salt, pool chlorine, industrial atmosphere | Use | Only if the exposure was declared in the enquiry and answered in the specification. Undeclared exposure is the most common silent exclusion |
| Colour difference between two batches | Coating batch variation, which needs a stated tolerance, a viewing distance and a viewing angle before it can be judged at all | Factory | Only with a retained reference sample and a stated method. Without both it is one opinion against another |
The honest version of "defect rate" is therefore a sentence, not a figure: the number of confirmed nonconformities against an approved list, per defined lot, counted at a named gate, by a named party, with an agreed remedy. Everything shorter than that is marketing.
Pre-shipment inspection verifies condition at origin. Site receipt inspection verifies condition after transit and handling, before installation. Public project-quality writing on the gap between them is blunt: procurement treats its job as finished when the pre-shipment report is clean, the installation contractor treats its job as beginning at installation, and nobody owns receiving. The interval between the two is commonly put at somewhere between four and sixteen weeks for international shipments, and it contains origin port handling, ocean transit, destination port handling, customs examination, inland transport and site offloading, each one an opportunity to introduce something the pre-shipment inspector could not possibly have seen.
Three things follow from that, and all three belong in the contract rather than in good intentions:
Where a genuine defect is in dispute, the practical advice published for buyers is consistent and worth following in this order. Inspect before shipment, because identifying a problem while the goods are still in the country of origin is far cheaper than any cross-border argument about it. Use an independent agency rather than supplier photographs. Record the unpacking continuously. Stop using the affected goods, keep them separate from sound stock and preserve a chain of custody, because goods that have been altered, repaired or consumed are very hard to prove anything about. Notify the seller immediately through a traceable channel: both Chinese contract law and the United Nations Convention on Contracts for the International Sale of Goods expect notice within a reasonable time after discovery, and delay weakens a position that was otherwise strong.
Two cautions from the same source matter more than the rest. A laboratory report commissioned unilaterally often carries limited weight where the supplier was not notified and had no chance to observe the sampling, so agree the testing institution, the sampling procedure and the applicable standard jointly, in advance. And a notarised inspection proves the condition of the goods on the day it was carried out; it does not by itself prove that the defect existed before shipment.
A defect is a departure from a declared threshold. That is why the most useful thing a supplier can give you is not a rate but a set of declared values you can inspect against. Here is exactly where ours stand.
| What we publish | What it covers | Why there is a gap, and what to ask |
|---|---|---|
| 6060-T66 profile, 1.8 to 2.5 mm wall, rising to 2.2 to 2.5 mm on the heavy sliding door | Alloy and wall thickness across the range | It is the same across series, so it does not differentiate one from another. Ask for the melt certificate and the measured readings, not the claim |
| PA66GF25 multi-chamber insulation strips, 34 to 44 mm on C116 | Thermal break material | Published strip widths exist only for some series. Ask for the width on the series you are buying |
| K values of 1.3 on AW90, 1.6 on C120 and 1.7 on C116 | Thermal performance of the casement and tilt-turn family | These are system values, not whole-window values and not certified values. No K value is published for the sliding series, because a sliding sash must keep a running clearance and can therefore only seal by contact rather than by compression. Unpublished is not worse, it means there is nothing comparable yet. Ask for a calculated value at your size, your glass build, and a named standard |
| 240-hour salt spray resistance and 5,000 N shear resistance, on AW90 | Coating and substrate corrosion screening, and the composite bond between strip and aluminium | Neither is a water, air or wind class, and neither transfers to another series. Do not read either one as a durability rating |
| 30,000-cycle testing on CMECH hardware | A bench result for a hardware item | It is not a service life, and handle cycles are not the same test as loaded sash cycles. Ask what was cycled, at what sash weight, in what configuration |
| Pulley load ratings of 400, 220 and 200 kg | Running gear capacity | Rated per sash, and governed by the glass. Maximum panel sizes are published against an assumed glass build, so changing the glass moves the panel limit |
| Frame wall widths: 90, 112 and 167, 116, 119, 126 and 182, 130, 155 mm | The depth that has to fit the wall | Frame depth is a series property, not a function of track count: the four-track 155 mm is shallower than the three-track 182 mm. Specify the number you are getting, in millimetres, on the order |
| Track geometry: 53 and 31 mm on the side-press window; 43 mm high and 8 mm low track; 45 mm high-ｄｒｏｐ bottom track with an optional 85 mm water barrier | Drainage and threshold behaviour | These are the numbers that decide whether water leaves or sits. Ask which combination your order carries, and get the drainage path drawn on the shop drawing rather than described |
The most relevant line in this article: no air permeability, watertightness or wind resistance class is published for any series. A defect is a departure from a declared threshold, so with no declared class, "the window leaks" is not a measurement, it is an argument. If your project needs a class, write the class, the standard and the test configuration into the enquiry, and ask for a report that names your configuration rather than a series name.
So we will not publish a defect rate for your project, and you should be cautious of anyone who publishes one without naming the gate. A rate belongs to a protocol, and the protocol is yours to write. What we can supply is the parameter set that makes the protocol writable, the drawings that make each clause checkable, and the batch records that make a claim decidable long after the container has gone.
| Clause | What to write | What goes wrong without it |
|---|---|---|
| 1. Approved defect list | Every defect you will count, classified critical, major or minor, with photographs, viewing distance and lighting conditions | The inspector counts one thing and you read another. This is where most disputes are born |
| 2. Sampling basis | Standard and edition, inspection level, and a separate AQL for each defect class, with critical set at zero acceptance | "AQL 2.5" alone is not a reproducible instruction. Two people can read it two ways and both be within their rights |
| 3. Lot definition | One lot per configuration. Separate glass builds, colours and panel counts are separate lots | A mixed lot hides the variant that is failing, and the average looks acceptable while one configuration is not |
| 4. Who draws the sample | The inspector selects independently from the finished lot, with all variants represented. The supplier does not pre-ｓｅｌｅｃｔ | Convenient cartons by the door are not a representative sample, and the whole exercise becomes theatre |
| 5. What is measured, not counted | Wall thickness against the mill certificate, coating thickness and batch, glass build, critical dimensions | Folded into the attribute count, these become pass-or-fail opinions. Kept separate as special controls, they become evidence |
| 6. When | A check during production, a pre-shipment check before loading, and a container loading check. Staging checks through production is common agency practice, and finds things a single end-of-line visit cannot | One visit at the end finds a finished lot, and the only available remedy is to reject it, which nobody wants and everybody delays |
| 7. Site receipt | A named owner, a deadline in days from arrival, inspection before unpacking and before installation, continuous video, exceptions on the delivery note | Transit damage is discovered during installation, by which time three parties have touched the goods and no one can be shown to have caused anything |
| 8. What happens on failure | Hold shipment, rework, sort, ｒｅｐｌａｃｅ, or accept with a written concession, and who carries the cost of the delay | The remedy gets negotiated after the failure, under schedule pressure, which is the worst possible moment to be negotiating |
| 9. Retained samples and batch records | A sealed reference sample per colour and coating batch, the profile heat number, the coating batch record and the glass build record | Without a retained reference, a colour or coating claim in year three is two people remembering different things |
| 10. The counting window | The date on which the product count closes and the installation and use period begins, and who signs it | Nobody can say when a number stopped, so every figure stays open to revision and every discussion restarts from zero |
| Situation | Why the protocol is the wrong purchase |
|---|---|
| One house, one or two openings | Building a measurement protocol costs more than the windows it would protect. Buy locally, from someone who will come back |
| You need one party to own measuring, installation and the warranty together | A factory can be excellent at its lot and cannot be the party that signs for installed work in your jurisdiction |
| The opening is not dimensioned or the structure is not built | There is nothing to inspect against. The defect you will get is a fit defect, and it is entirely predictable |
| You need a spare part the same day | The rate is irrelevant; the stock location is the specification. Buy the critical spares into the original container and the question disappears |
| The programme is shorter than production plus sea transit | The defect you will experience is a schedule defect. No inspection clause has ever recovered a missed handover |
| Nobody in your organisation owns the quality file | The count will still happen, it will just happen after handover, in the wrong currency, with no records and no remedy |
| Ask | A good answer names | Sounds reasonable, tells you nothing |
|---|---|---|
| What is your defect rate? | The gate, the defect list, the denominator, the date and the remedy | "Below one percent" |
| What sampling plan do you use? | Standard and edition, inspection level, AQL per class, and who chooses the units | "We work to AQL 2.5" |
| Can I see the inspection report? | Lot identity, how the sample was drawn, defect classes, counts, and the disposition | A folder of photographs |
| What is your air, water and wind class? | Standard, class, the tested configuration, and the laboratory | "It meets international standards" |
| Is water in the track a defect? | The drainage path, where the inner upstand is, and where the weep holes exit | "Our windows do not leak" |
| What happens if the lot fails? | Rework, sort, ｒｅｐｌａｃｅ or concession, and who pays for the delay | "We will take care of it" |
| Who loads the container, and is it recorded? | Method, materials, stillage type, and loading photographs tied to the container number | "Professional export packing" |
| What exactly is the glass build? | Full build, spacer type, gas fill, coating, and the panel size it was rated against | "Standard double glazing" |
| What spares ship with the order? | Part numbers, quantities, which travel in the original container, and the reorder route | "Always available" |
| Who owns site receipt? | A named role and a deadline in days | "The installer will handle it" |
We do not publish one, and we think you should treat a bare figure from anyone as incomplete. What we can publish is the parameter set that lets you define one: alloy and wall thickness with mill certificates, insulation strip material and width, thermal values by series, corrosion and shear results for the series they belong to, hardware cycle testing, per-sash load ratings, frame depths, and the track geometry that governs drainage. Give us the gate, the list and the denominator, and the number becomes measurable.
No. That is the single most common misreading in sourcing. AQL 2.5 is the worst process average that will usually still be accepted across a continuing series of lots. The actual decision comes from the sample size and the acceptance and rejection numbers of the plan you selected, and the worked examples above show a truly 2.5 percent defective lot passing about 99 percent of the time, with a 5 percent defective lot still passing a majority of the time.
No. It covered one sample, on one day, in one location, before transit and before installation. It is a decision rule with a known error rate, not a description of your shipment. The two gates that usually matter more are the one during production, where a remedy is still cheap, and the one at site receipt, where transit damage is still attributable.
Usually no. A sliding frame is built as a managed drainage system: the outer seal reduces what gets in but is not expected to stop it under wind pressure, and what enters the outer track leaves through the weep holes. Water past the inner upstand is the failure, and the first thing to check is whether the weep holes have been painted or caulked over. Sealing the visible gap is the most common wrong repair, because it closes the path that was working.
The honest answer is that the number moves more with the counting window than with the flag. The worst published figures come from market-wide sampling that includes price tiers and producers no project buyer would ever approach. The middle figures come from installed projects, where a large share of what is counted happened on site after delivery. The best figures come from the factory gate, where the counting stops earliest. If you want a number that describes your order, you have to define where yours stops.
Whoever your contract says. If it is silent, it will be settled by whoever is more invested in ending the conversation, which is a poor way to run a project. Write it in, and note two things while you are there: replacement parts often ship on terms that leave import duties with the buyer, and the invoice should state no commercial value, for customs purposes only. Better still, put the critical spares into the original container and skip the argument entirely.
