Quality control

Quality control is a sampling process, not a counting one. That single fact explains most of what surprises people about it — including why a passed lot can still contain a defective item, and why that is not a failure of the process.

The approved sample is the standard

Everything is judged against a physical reference that both sides have signed off. Not a mockup, not a description — an actual made item.

This is why sampling is not a formality to be rushed. Whatever you approve becomes the definition of correct. If the sample has a print positioned two centimetres low and you approve it, the run is correct at two centimetres low, and you have no basis for rejecting it.

Approve deliberately. Measure the sample rather than eyeballing it, and write down what you measured.

How sampling works

Inspecting every unit sounds thorough and is not. It costs more than the defects it finds and it is less reliable — human attention degrades quickly under repetition, so the hundredth item is checked worse than the first.

Instead, a statistically determined sample is drawn from the lot, sized by the lot’s size and the inspection level. The sample is inspected fully. The number of defects found determines whether the whole lot passes.

The threshold is the Acceptable Quality Limit. It is a limit on the sample, from which a confidence about the lot is inferred. A pass does not mean zero defects exist; it means the defect rate is statistically likely to be below the agreed level.

That is the honest position, and it is worth internalising before you find one bad shirt in a passed delivery and conclude the inspection failed.

Defect classes

Three, and they carry very different weight:

Critical — a defect that makes the product unsafe or unlawful. A needle fragment, a choking hazard, a missing legally required label. Tolerance is zero. One is enough to stop a shipment.

Major — a defect a customer would notice and reject. Wrong colour, print misaligned beyond tolerance, a hole, a broken seam, wrong size in the bag, print peeling. These drive returns, so they carry the tighter of the two commercial limits.

Minor — a departure from specification a customer would probably accept. A loose thread, a very slight shade variation, marginally off-centre print within tolerance. Allowed at a higher rate, because eliminating them entirely costs more than they cost you.

Agreeing where the line sits between major and minor, before production, is one of the highest-value conversations available. “Slightly off-centre” is a judgement, and it is much cheaper to define it as a measurement than to argue about it afterwards.

What gets checked

Measurements against the size chart, at defined points — chest, body length, sleeve — with a stated tolerance. Garments are sewn from cut fabric and will never be exact; the tolerance is the point.

Print or embroidery placement, measured from a fixed reference such as the shoulder seam, again with a tolerance.

Colour, against the approved sample under controlled lighting. Two colours can match under one light source and differ under another — an effect called metamerism — which is why lighting is specified rather than assumed.

Print durability. Adhesion checked by tape or stretch. Cure verified, because an under-cured print looks perfect and fails on the first wash.

Construction. Seams, stitch density, loose threads, needle holes, skipped stitches.

Labelling. Correct content, correct size, correct placement, legally required information present and legible. See the compliance guide.

Packing. Right item, right size, right quantity, right carton, right markings.

When a lot fails

Four options, and the choice is commercial rather than technical:

• Rework the defects, where the defect is fixable.
• Sort at 100% to pull defective units out, where it is not.
• Remake, where the defect is systemic.
• Accept with a concession — a documented, deliberate decision to ship known-imperfect goods.

The last one is legitimate and is also where discipline usually fails. Under deadline pressure a concession gets granted informally by whoever is available. It should be a named decision, recorded, with the reason attached, because the person who has to answer for it later will not be the person who granted it.

What you should insist on

• An approved physical sample before production, not a photograph
• Written tolerances for measurement and placement
• An agreed AQL and inspection level, stated before the run
• An inspection report with the sample size, defects found and their classification
• A named person with authority to accept, reject or grant a concession

None of that is exotic and all of it is cheap to agree beforehand. All of it is expensive to argue about once the goods exist.