Quality control is usually presented as a set of inspections. It is more useful as four points on a production timeline where a particular class of failure can still be caught — and after which the same failure gets dramatically more expensive to fix.
Why the timing matters more than the check
All four use much the same tools: a tape measure, a specification, an approved reference sample, a list of what counts as a fault. What separates them is when they happen — and therefore what can still be done about a finding.
A wrong fabric found at sample approval costs a revision round. Found at final inspection it costs the fabric, the cutting, the sewing and the shipping date. The inspection did not change; the cost of the finding did. That is the argument for pushing checks earlier, and the honest limit on the last one: a pre-shipment inspection cannot fix a materials decision made eight weeks earlier. It can only tell you it happened.
The four checks at a glance
| Check | Catches | Cannot catch | Cost of a failure found here |
|---|---|---|---|
| 1. Sample and pre-production approval | A misread design, an unachievable construction, the wrong material or trim, measurements the pattern cannot hold. | Anything about consistency. One good unit says nothing about the thousandth. | A revision round and some calendar. No bulk material is committed. |
| 2. Incoming materials and pre-production check | Substituted or off-standard fabric, components and trims; the wrong colour lot; a line set up against the wrong reference. | Faults that develop later in sewing, assembly, finishing or packing. | Replaced material and a delayed start. Cheap against what it prevents. |
| 3. During-production inspection | A fault repeating across units, drifting measurements, a machine or operator problem, a mid-run substitution. | Anything finished and packed before the visit, or appearing only in the last operations. | Rework on work in progress, plus the units already made wrong. |
| 4. Pre-shipment final inspection | Finished goods against the specification and approved sample on a defined sampling plan, plus packing, labelling and quantity. | Any decision made upstream. It reports the outcome; it cannot change the input. | The whole order. Rework, sort, discount, part-ship or reject — all late. |
Check 1 — Sample and pre-production approval
What is checked: whether a physical unit made on the intended materials and process matches the specification, and whether the specification is achievable at all. When: after development rounds, before bulk materials are committed. By whom: the buyer, against a submission from the factory.
A failure here means the design, the specification or the material choice has to change — the cheapest change you will ever make. The output is an approved reference that every later check compares against; what a golden sample is covers what it has to be. The limit of this check is that it proves capability, not consistency.
Check 2 — Incoming materials and the pre-production check
What is checked: that the fabric, components, trims, labels and packaging arriving at the factory are the approved references — right supplier, right specification, right colour lot — and that the line has been set up against the approved sample and current measurement chart, often with a small pilot run first. When: materials in, line about to start. By whom: the factory's incoming inspection, verified by the buyer or an appointed inspector.
This is the most valuable failure to find, because material faults are unfixable later and affect every unit. It is also the check brands skip most often, because it happens out of sight at the moment everyone is relieved production is starting. A substituted trim or a colour lot from a different dye batch is trivial to correct now, impossible after cutting.
Check 3 — During-production inspection
What is checked: units taken from the line while it is running, measured against the same specification and sample, plus the process itself — how operators are working, what the factory's own checks are catching, how much rework is happening. When: once a meaningful share of the order is complete, with enough left to act on. By whom: the factory continuously, the buyer or a third-party inspector at least once.
The point is not the units inspected but the pattern. A single defect is noise. The same defect in the same place, unit after unit, is a cause — a machine setting, a misread instruction, a component out of specification — and it continues for the rest of the run until someone stops it.
Check 4 — Pre-shipment final inspection
What is checked: finished, packed goods against the specification and the approved sample on an agreed sampling plan — measurements against the stated tolerances, appearance and workmanship, materials and trims, function where the product does something, labelling, barcodes, packaging, and the quantity actually there. When: when the order is complete or nearly so, before it leaves the factory. By whom: the buyer or an appointed inspector, on a sample drawn at random from the packed lot.
This is the last point at which "no" is cheap — once goods are on the water the leverage has gone. But be honest about what it is. A final inspection reports the state of the goods; it cannot undo a materials decision, a pattern error or a process fault. A brand relying on it as its only check has arranged to discover its problems at the most expensive possible moment.
AQL: an acceptance rule, not a permitted level of bad product
- Acceptance quality limit (AQL)
- The value that indexes a sampling scheme — not a defect allowance. In ISO 2859-1's own terms, the scheme uses the prospect of a lot not being accepted to press the producer into holding a process average at least as good as the specified limit, while capping the buyer's risk of accepting an occasional poor lot.
You will often hear AQL described as "the acceptable percentage of defects". That is the wrong way round, and the error is not pedantic: the number is not permission to ship a share of bad units. It is an input that, with the lot size and an inspection level, produces a sampling plan.
The mechanism has four parts, none of them a share of your order you agreed to accept:
- Faults are classified into agreed classes by seriousness, the most serious carrying the smallest limit, and the standard reserves special provisions for classes designated critical. Which fault sits in which class is your decision, not the standard's.
- A lot is formed and an inspection level chosen. With the lot size, the level yields a sample size code letter.
- The code letter and the limit for each class give a sampling plan: how many units to draw, and the acceptance and rejection numbers. Single, double and multiple plans exist.
- Across a continuing series of lots from one producer and process, switching rules move inspection between normal, tightened and reduced — deteriorating quality met with more inspection, consistent quality with less.
Two consequences matter more than the arithmetic. First, all of it has to be agreed before the inspection, in writing: the classes, the limit for each, the inspection level, the plan, and what happens to a lot that is not accepted. Settled after a failed inspection, it is a negotiation. Second, the scheme is built for a continuing series of lots from one producer using one production process — ISO 2859-1 says so explicitly — so a first order from a new factory is the case switching rules cannot help with.
It is worth knowing which document you are pointing at. The current edition is ISO 2859-1:2026, the third, published in January 2026; it replaced ISO 2859-1:1999, now withdrawn. The United States counterpart is ASQ/ANSI Z1.4-2003 (R2018), descended from MIL-STD-105E — a military standard the US Department of Defense stopped supporting in 1995, recommending Z1.4 in its place. "AQL" read "acceptable quality level" in those military standards, which is where the misdescription started.
What an inspection report has to contain to be usable
An inspection is only as good as the record it leaves; most disagreements are lost on the record rather than on the goods.
- Minimum contents of a usable report
- The reference used: specification version, measurement chart version, and the approved sample.
- The sampling plan applied — lot size, inspection level, the limit for each class, sample size, acceptance and rejection numbers.
- Measurements per point per unit, as numbers rather than a pass mark.
- Findings classified into the agreed classes and counted, each photographed.
- Photographs of the goods as packed — cartons, labels, barcodes — not only of defects.
- Quantity found packed against quantity ordered.
- The accept or not-accept conclusion, who reached it, and the date.
Photographs are the part brands under-specify. A close-up of a stain proves a stain exists somewhere; a photograph showing the fault, its position and something for scale is evidence. And a carton photograph is the only record of what was in the box.
What happens when a check fails
A finding needs two separate answers, and confusing them is why quality conversations turn into arguments. Disposition is what happens to this product — rework, sort and ship the conforming part, accept with a documented concession, or do not accept. It is a commercial decision, it belongs to a named person, and it is recorded with its reason.
Correction is what changes so it does not recur: a root cause, a change to whatever allowed it — an ambiguous specification, an unachievable tolerance, a component source, the inspection plan itself — and a check on the next run that the change worked. A failure that produces only a disposition produces the same failure next season. Reducing supplier and quality risk before production covers how that record turns a failure into a change rather than a dispute.
How Library of Trade runs these checks
The four layers map onto stages we manage: sample approval in Sampling, materials and line set-up before Production, in-line checks during it, and final inspection in Quality before a shipment is approved. What we make structural is the part that usually leaks — the reference and the acceptance rule are set when the specification and the sample are approved, not invented at the inspection. AI compares the records; the accept-or-reject decision goes to a person.