400 mattress covers carried fully compliant wash labels. Three were washed to 40 °C and two came apart at the seam, and after three washes the cover fabric and its zip binding were 26 millimetres out of step.

Four hundred covers and what was promised

A European home-textile brand ordered 400 mattress covers for a serviced-apartment project: 1,600 by 2,000 mm, a quilted 65/35 poly-cotton panel, a zip on three sides, a care label in symbols, machine washable at 40 °C, colour fastness to grade 4 or better, and no more than 3% panel shrinkage. The unit price was $28, so the order was $11,200. A 30% deposit of $3,360 was paid. The $7,840 balance was not.

The order named the label and the wash temperature. It did not name how many cycles the cover had to survive, it did not name a seam strength, and it did not say that the binding around the zip and the cover panel had to shrink by the same amount. Those three omissions are the whole case.

A certificate came with the shipment stating that the wash test had been passed at 40 °C. One cycle. Everything on paper was complete, and everything on paper was about the fabric. Not one of the documents described the cover as an assembly of three fabrics that had to agree with each other after they got wet.

The label check passed on all four hundred

At the factory the label was checked properly, because a label is cheap to check and expensive to get wrong. All 400 covers were unpacked and their care labels read against GB/T 8685 and GB 5296.4-2012: size and specification, fibre composition and content, and the maintenance method, all present, printed with the correct symbols, legible, and attached as a durable label rather than a hang tag.

The general attribute inspection ran on the standard sample. A lot of 400 units is code letter H, general inspection level II, so n = 50, and AQL 2.5 gives an accept number of 3 and a reject number of 4. The label check was still run on all 400 rather than on 50, because a wrong fibre-composition or care symbol is a compliance defect, and for a European buyer the composition label is a legal requirement rather than a preference.

Three lines went into the report. Care labels: 400 of 400, conform. General inspection: pass at AQL 2.5. Supplier wash certificate: accepted. The buyer had three documents in agreement and a cover that would not survive its own care instructions.

What a care label claims, and what it does not

Reading a wash label is a fixed exercise and it is worth doing precisely. The number in the tub is the maximum water temperature the article may be washed at, and it was 40 here. A single line under the tub is the mild process, two lines is the very mild process. A crossed triangle means no chlorine bleach. A square with a vertical line means line drying, a circle inside the square is tumble drying, and any dot on it caps the heat. Dots on the iron cap the sole plate temperature, and the letters P, F and W in a circle are the professional dry-clean and wet-clean routes. On this order every one of those symbols was right.

What none of them says is how many times the article can be washed. The symbols in GB/T 8685, drawn from ISO 3758, state the most severe care process the product may be exposed to without damage. They are a ceiling written for the person doing the washing, not a floor written for the person who made the cover. There is no symbol for a cycle count, and no symbol for a seam. A label that is lawful and truthful can still be attached to a cover that opens at the third cycle, because the truth it is required to state is about care limits, not about how long a sewn article lasts.

GB 5296.4-2012 supplies the other half of the obligation: the durable label must stay attached and legible for the product's service life, and the information on it must be truthful. So the label did have to survive the wash, and it did. Nobody asked the cover to.

Three covers and one washing programme

The wash test was run on three covers, pulled from three different cartons and destroyed in the process: three covers at $28 each, so $84 of stock written off. The programme was the domestic washing procedure of GB/T 8629 at 40 °C with tumble drying, repeated, with the covers measured and examined after every cycle.

Nothing visible happened on the first wash, which is exactly why a single-cycle certificate was worthless on this order.

The third cover is the reason this case is worth reading at all. It had not failed, and it carried a 14 mm pucker at the same corner where the other two ruptured or came apart. The failure is progressive. A test that stops at cycle 3 can certify a cover that is halfway to the same failure.

Three fabrics, three shrinkage rates

The three covers were then measured for dimensional change to GB/T 8630, and the covers were opened up so each component could be measured on its own. That last step is the one that matters. Shrinkage is a property of a fabric. The failure was a property of an assembly.

None of these three fabrics is wrong. The quilted panel shrank the most because the quilting stitches hold it under tension and the first wash relaxes that tension. The flat twill binding was already relaxed at the mill. The polyester chain tape sits between them. The buyer had one limit for the panel, 3%, and 2.8% passed it.

On the long edge the panel lost 38.0 mm of length and the binding strip lost only 12.0 mm, so the panel finished 26.0 mm shorter than the strip it was sewn to. A sewn seam is a fixed length, so that 26.0 mm cannot simply disappear: it has to be taken up somewhere, and the panel, thinner and quilted, gave way along the stitch line instead of the thread snapping. On the short edge the difference was 35.2 mm, and it concentrated at the corner where the strip turns 90 degrees, which is where the 45 mm rupture appeared on the first cover.

The zip failed for the mirror-image reason. The chain tape lost 28.0 mm over 2,000 mm while the strip it was sewn to lost only 12.0 mm, so the chain ended up 16.0 mm too short for the strip it had to close. The surplus strip bunched at the corner and drove the slider out of the chain, separating 25 mm of it on the second cover. Two fabrics that disagree by 26.0 mm, and a third that disagrees with the second by 16.0 mm. Nothing in the order carried a number for any of it.

Recording the failure without overstating it

The report had to keep three questions apart, because buyers tend to merge them: does the defect exist, how far does it extend, and what causes it. The answers here were yes, not measurable, and measured.

Existence was easy. The lot fails. A failure traced to the construction of the cover is systemic rather than sporadic, and the inspector confirmed that from the factory's own records instead of assuming it: every one of the 400 covers had been cut from one panel roll lot and one trim lot, with the roll tickets and cutting records on file. A sporadic defect gets counted. A systemic one is not a count of bad units, it is a lot that fails.

Extent was a statistical statement, and a weak one. Two failures in 3 washed covers is a point estimate of 67%, and the 95% Clopper-Pearson interval around it runs from 9.4% to 99.2%. On 400 covers that is anywhere from at least 38 covers affected to at most 396. Three samples cannot narrow that further, and a report should not pretend otherwise.

Cause was the strongest part of the file, because it was measured rather than inferred: the shrinkage figures for panel, strip and chain tape, the 26.0 mm and 16.0 mm differences, the rupture at the corner, the separated chain. A report that stops at 3 covers washed and 2 failed leaves the factory with an argument. A report that ends with 26.0 mm leaves it with a problem to fix.

Why three samples cannot settle four hundred covers

The temptation here is to reach for AQL, because AQL is the language buyers are given for deciding how many units to look at. It does not answer this question, and the reason is worth stating plainly. AQL estimates the number of defects present in a lot at a point in time, for attributes you have chosen. Wash durability is a cumulative property that has to be demonstrated by running the test, and the test destroys the unit.

So the arithmetic runs the wrong way. To hold the bad covers in the lot under 10% with one-sided 95% confidence, 36 covers have to be washed; under 5%, 72; under 2%, 183, which is 46% of the shipment destroyed and written off. Fifty covers, the code letter H sample for this lot, would only bring the bound down to 7.1%.

That asymmetry is what the specification has to absorb. A wash test on three covers can condemn the lot, but it can never clear it. Even three clean washes would only exclude the bad covers above 70.8% of the lot, which is 283 of 400. A test that can only fail does not belong in the acceptance clause as a pass or fail gate on the buyer's sample. It belongs as a documented type test the factory runs and the buyer's inspector verifies.

A mattress inspection Guangdong buyer is often told a failure rate is a countable number of bad units. On a systemic failure it is not.

Writing the wash clause so it can be tested

The clause failed because it described a care regime instead of a performance. A regime is a set of instructions. A performance is something that can be measured and given a number, and only the second can be checked by somebody who is not the factory. What to write instead:

• Care label to GB/T 8685 on a durable label per GB 5296.4-2012, carrying size, fibre composition and content, and the maintenance method, and remaining legible and unchanged after the wash cycle in clause 2.
• Wash test: five cycles at 40 °C under the GB/T 8629 domestic washing procedure with tumble drying. The number of cycles has to be written into the order, because no symbol carries it.
• Seam: no rupture of the cover panel at the placket seam after the five cycles; where a strength figure is required, ISO 13935-1 seam strength of at least 40 N after the same five cycles.
• Differential shrinkage: no more than 1.0 percentage point between the cover panel and each trim, binding and zip tape after the five cycles.
• Panel shrinkage: not more than 3% warp and weft measured to GB/T 8630 after the five cycles.
• Colour fastness after the five cycles: GB/T 3921 water grade 4 or better, GB/T 3922 perspiration grade 4 or better, GB/T 3920 dry rubbing grade 4 or better, tested on the production fabric rather than on a sample book.
• Zip: chain and placket edge lengths within 5 mm of each other at both ends before and after the five cycles, and the slider closes and opens 20 times after the cycles without leaving the chain.
• Type test: five covers from the production lot, not from the retained sample, washed by the factory at its own cost with the readings recorded, and the balance released only when the buyer's inspector confirms the type-test record and the re-inspection passes.

Where the balance ended up

The buyer held the $7,840 balance and did not release it. The factory reworked all 400 covers with components from one shrinkage family, either a pre-relaxed panel or a matched binding and chain, at $5.50 per cover, so $2,200, and 12 days. The single-cycle certificate was replaced with a five-cycle type-test record with the readings attached, and five covers from the reworked lot were washed to five cycles as a condition of acceptance.

The check on the reworked lot ran at the Full Verification level: two man-days at $299, so $598. That covers reading the 400 labels, verifying the five-cycle type-test record against readings taken on five covers at the factory, measuring shrinkage on panel, binding and chain tape, and verifying the third-party colour fastness report line by line. $598 measured against $2,200 of rework and $7,840 of balance.

And the honest limit of that fee: we do not run laundry. The wash cycles belong to the factory or to a laboratory, and our job is to verify the record, measure the components and check that the numbers in the certificate describe the covers in the cartons.

The check that mattered was not a label check

Three documents agreed with each other on this order, the care label, the supplier's wash certificate and the approved sample, and none of them described the thing that failed. The failure was in the relationship between three fabrics, and a relationship cannot be read off a label. Each part has to be measured on its own, after the water has been to it.

A lawful, legible and truthful label is not a warranty. It says what the consumer may do. How many times a sewn cover survives that treatment is a number somebody has to write down, measure and stand behind, and if nobody writes it down, the label wins by default and the buyer discovers it at cycle 3 with 400 covers and no recourse.