An inspection report reading PASS and a seller account in good standing feel like the same fact. They are not, and the distance between them is arithmetic rather than opinion. AQL 2.5 is a rule for deciding whether to accept a shipment, and the word in the middle of it is limit: a batch can pass while containing defective units, because that is what the standard was built to do. A marketplace measures something else — the share of your orders that produce a complaint. Here is how the two numbers relate, and where the connection breaks down on a small order.
The short answer
- An AQL 2.5 inspection is a sampling decision about whether to accept a lot. A pass means the sample did not exceed the acceptance number — not a clean bill of health, and not an estimate of the batch's true defect rate.
- Marketplaces measure something different: the proportion of orders generating a defect signal — negative feedback, an A-to-z style claim, a service chargeback. The largest marketplace publicly documents a target under 1%. Check the current figure and its definition on your own account's policy pages rather than trusting this page.
- So the two are not comparable. A defect level that clears an AQL acceptance comfortably can still sit well above the level at which an account gets into trouble: one is a limit for accepting a shipment, the other a ceiling on customer-visible failures.
- On a 300-unit order the inspected sample is around fifty units, and a sample that size cannot resolve a low defect rate. That is the honest limitation of pre-shipment inspection at small volumes, ours included, and it is why wear-and-wash testing belongs at sample stage.
Two numbers, two different questions
An AQL inspection answers one question: should this shipment be accepted? It is a lot-acceptance rule, and its own name concedes the rest — an acceptable quality limit. A lot can be accepted while containing defective units. That is not the method failing; it is the method working as designed, bounding risk at a stated level instead of opening every polybag. How the sampling itself works is covered in our AQL 2.5 explainer. The word that matters here is limit.
A marketplace measures something structurally different: not units in a lot but orders in a rolling window, and the share of them that generate a defect signal — a negative review, an A-to-z style guarantee claim, a service chargeback. The largest marketplace publicly documents a target below 1%. Treat that as a pointer rather than a fact about your business: definitions, windows and consequences differ by marketplace and get revised, and the only version that governs you is the one currently on your own account's policy pages.
The mismatch is not subtle. Different denominators — units in a shipment against orders in a window. Different definitions — a defect an inspector sees on a table against a customer annoyed enough to act. “The batch passed AQL 2.5” and “this batch is safe for my account” are two claims, and nobody in your supply chain, us included, should let you hear them as one.
The arithmetic, done out loud
Take round numbers, labelled clearly as an illustration rather than a prediction. You order 300 units and, across a quarter, every one sells to a real person. Suppose the batch is genuinely running at two major defects per hundred — a level a standard AQL 2.5 plan is designed to accept without argument. Two per hundred across 300 units is six garments: six people receive a legging that goes sheer at the squat, or a bra whose print has cracked.
Now convert those six into the number a marketplace watches. If all six act — a review, a claim, a chargeback — that is six defect signals against 300 orders, or 2%: twice the publicly documented target. If only half bother, three against 300 is 1%, sitting on the line with nothing left for the failures no factory caused. A late delivery, a crushed parcel, a size that did not fit land in the same numerator. It is not a quality metric; quality is part of it.
Small accounts get the worse end of this, because the percentage runs over orders in a window: a seller with a hundred orders in that window needs one claim to be at 1%. And here is the honesty the example needs — nothing in the inspection told you the rate was two per hundred. That was an assumption, inserted to make the arithmetic visible. An accepted batch might be running at two in a thousand, or three in a hundred. The pass does not say which.
What a pass proves, and what fifty units cannot
For a 300-unit lot at the normal inspection level, the published sampling tables put the sample at fifty units, and for major defects at AQL 2.5 the acceptance number is three. So “the batch passed” expands into a much narrower sentence: an inspector drew fifty units at random and found three or fewer major defects. Notice what that permits — a lot accepted with three defective units in a fifty-unit sample, six per hundred inside the sample itself.
The resolution problem, in numbers. Six defective units hidden among 300: draw fifty at random and there is roughly a one-in-three chance the sample contains none of them. Push the true rate to five per hundred — twice the AQL number — and fifty units still return three or fewer defects, and therefore a pass, about three times in four. Even a batch at ten per hundred slips through around one time in five. The tables are explicit: a plan is defined by how likely it is to accept lots at various true rates, not by a promise.
The plain conclusion, which almost nobody in this trade says out loud: a small sample cannot resolve a low defect rate. On a 300-unit order you are not measuring your defect rate, you are checking that it is not catastrophic. That is worth doing — it catches the wrong fabric, the mis-set machine, another order's labels, the systematic error running through everything. It is not evidence that six customers will not receive a bad garment, and the answer is not to trust the report harder.
The failures that only appear after wear and wash
A sourcing practitioner put the second problem bluntly in a public thread: a great many defects “manifest themselves in wear and tear”, and would never be found in a normal QC inspection, where a zip is opened and closed a couple of times. An inspection is a snapshot of a garment folded in a polybag. Anything needing time, body heat, sweat, friction or a washing machine to appear sits outside what that snapshot can contain. That is structural, not carelessness.
For activewear, the category is close to a list of the things that generate one-star reviews. Pilling on brushed faces. Recovery loss, so a legging that fitted in week one bags at the knee by week four. Print and heat-transfer logos cracking after washes. Dye transfer onto a pale sofa or car seat. Seam slippage at the gusset or bra band under stretch. Every one can be present in a batch that passed final inspection with a clean report, because none was visible on the table.
The only place they can still be caught is the sample stage, so put them there deliberately. Wash the pre-production sample repeatedly at the temperature your customer will really use, and measure the key dimensions before and after. Stretch, hold, release, measure the growth. Rub a damp garment against white cloth. Stretch the printed area and look at the edges. A formal test panel does this properly at material approval; a brand with 300 units and a launch date can still run a crude version in a week, and crude beats absent.
Write the defect list yourself, and tighten it selectively
An inspection is only as specific as the list the inspector walks in with. Accept a generic apparel checklist and the word “workmanship” does an enormous amount of undefined work — in practice, a reasonable general standard applied to a product category with unusual failure modes. Before anyone goes in, write down what a major defect is for your product, in terms a person who has never worn your brand can check in a fixed number of seconds.
For activewear that means naming things rather than gesturing at them. Opacity: checked in a stated stretch position under stated lighting, against a physical pass/fail reference instead of an opinion. Recovery: a permitted growth percentage after a defined stretch and hold, not “good stretch”. Print and logo adhesion: a specified stretch or peel action, with the acceptable result described. Seam behaviour: under stretch, at the joins that carry load. Colour: against the approved reference, same lighting, every lot.
Demanding a tighter AQL across the board is the obvious answer, and usually the wrong one. A tighter plan means more units inspected, more re-inspections, longer lead times and a supply chain that prices all of it — and it buys little on defects a customer will never see. The sensible version is asymmetric: tighten where a defect is customer-visible and account-threatening, consider a 100% check on the one or two characteristics that matter most, and accept an ordinary line on cosmetic details nobody will photograph.
The reorder, and the invoice you do not have yet
Scale-up is the risk event, not the safe one. One seller's public account: the first order came through fine, then “the second (bigger) order had a 20% defect rate”, producing hundreds of returns and an account suspension within weeks. One reported experience, not a statistic — but the shape is familiar. A bigger order is not the same order repeated: a different line, often a second facility when capacity is short, a new fabric and dye lot, sometimes a substituted trim, and usually rushed.
The metric arithmetic turns at the same moment, because a reorder pushes far more orders into the rolling window in a shorter time; a first order that sold through slowly may never have moved the percentage at all. So treat the reorder as the higher-risk event: confirm the sealed reference physically exists at whichever line is running, re-approve the fabric and dye lot rather than assuming continuity, and inspect the scale-up at least as hard as the first order. The instinct to relax is precisely backwards.
The second trap is documentary. One seller described about a third of the inventory for a product, sitting at a fulfilment centre for two months, being flagged as defective — and the sting was not the defect. Without a proper commercial invoice from the manufacturer, the units could be disposed of with no reimbursement at all. A brand buying 300 units often cannot produce one: it bought through a platform storefront whose registered selling entity is not the factory on the carton, paid an account in a third name, and holds a proforma or a chat thread.
That is created at purchase-order time, months before any defect exists — cheap to fix then, close to unfixable afterwards. Keep the commercial invoice from the entity you actually bought from, in your company's name, from the first order rather than from the first dispute, with the packing list and inspection report beside it. Marketplaces publish document requirements of roughly this shape and revise them, so read the current version. Whether a supplier can issue that invoice at all is worth asking before you pay one, us included.
Where our own inspection stops
What SEAMDANCE runs is a final inspection to AQL 2.5 for major defects before shipment, carried out on the buyer's side rather than by the producing facility, against the approved reference. That is a real gate: it catches the systematic failure — the wrong fabric, the mis-graded run, another order's labels — and a failed lot does not become a shipped lot because a date is tight. We coordinate independent partner facilities, mills and dye houses rather than owning production, and inspection is where our name sits on the outcome.
Its limits apply to us exactly as to anyone. It is a sampling gate, so on a 300-unit order it cannot resolve a low defect rate — the fifty-unit arithmetic above is our arithmetic too. It is a snapshot at packing, so it cannot see the pilling, recovery loss, print cracking or dye transfer a customer's fourth wash will find. And it accepts a lot against a limit, so it will sometimes correctly pass a batch that later produces complaints. None of that argues for skipping it. It argues against hearing “passed AQL 2.5” as “your account is safe”.
If your account survival depends on near-zero customer-visible defects, the conversation worth having is before the order: a tighter standard on the characteristics that threaten the account, wider or 100% checking on one or two of them, and wash-and-wear testing at sample stage. Each costs money and time, which is exactly why it should be a decision you make rather than an assumption you carry. We would far rather set that expectation up front than defend a correctly passed batch afterwards.
This is general information rather than advice about your account. Marketplace policies, thresholds and document requirements change, and the version that governs you is the current one on your own seller policy pages.