FAQ · incoming inspection · in-line checks · ISO 2859-1 AQL · defect classes · credits and remakes
A written procedure tells a factory what to check. This one explains what each check actually proves, what it misses, and how a failed batch gets resolved. Written for the person who has to sign the inspection off, not for a procurement manual.
Merch production in Armenia is a short, hands-on process. A 200-piece screen print run on cotton tees takes about two days once the plates are burned. A 1,500-piece laser run takes five. Nothing here runs long enough for a defect to hide for a month, which is good news, because the same short cycle means nobody is looking at your goods on a schedule unless someone wrote one down.
Three inspection moments cover the cycle. Each one is cheap relative to the batch and each one catches a different family of problems. Buyers who arrange only the last of them usually discover the first two did their job too late.
Blank tees, ceramic mugs, glass tumblers, notebooks and tote canvas arrive from suppliers before anything is printed on them. This is where you confirm the fabric weight against the supplier's card, count units per colour, look for shade variation between cartons, and check that size labels are right. For drinkware, this is where you find chipped rims and hairline glaze cracks, before they get printed and packed.
What it cannot catch is anything about your artwork. A perfectly delivered blank can still be the wrong colour, and that will only show up later against the Pantone chip you approved.
After the first run of a print job stops, before the operator runs the rest. We set one piece aside and check registration, placement, ink coverage against the approved colour standard, and cure temperature on screen work. On embroidery we check stitch density and the backing, and on laser we check engraving depth. It takes fifteen minutes and it stops a 1,500-piece run from being 1,500 identical mistakes.
The judgement here is about tolerance. Screen printing on a knit will never hold a registration tighter than roughly half a millimetre without the garment distorting under the squeegee, so a check against a zero-tolerance rule is a check that always fails and tells you nothing.
Once packing is finished, a sample is drawn at random from the packed cartons and checked against the specification: correct item, correct print, correct colour, no damage, correct labelling. This is the only moment when the goods exist as the buyer will receive them, and it is the moment that supports or refuses a claim later.
It is also the moment most often skipped on small runs, which is a mistake of the opposite kind. A 50-piece order does not need a statistical sample. It needs someone to look at all 50.
Four jobs sit inside this process. They have different owners, different evidence and different consequences, so it is worth keeping them apart rather than calling all of them quality control.
Weigh a sample of the knit against the stated GSM, count units per colour, compare cartons under daylight for shade drift, and set aside chipped or cracked items. Runs when the supplier's truck arrives at the Yerevan workshop or the Vanadzor unit. Takes about 20 minutes for a 500-piece delivery.
One garment or mug comes off the press and stops. Registration, placement relative to the print area, colour against the approved reference, and cure temperature for screen work get recorded on a sheet the buyer can sign. Nobody runs the remaining pieces until that sheet is signed or waived.
A random sample drawn from packed cartons, inspected to ISO 2859-1 at General Inspection Level II. The result is a decision, accept or reject, tied to a lot number and a date. This is the record a buyer keeps if a claim is raised after delivery.
When a check fails, someone decides between rework, reprint, downgrade to seconds or rejection, and writes that decision down with photos. The commercial terms attached to that decision come from the purchase order, which is why the purchase order is the document that actually protects you here.
Inspecting every item in a large lot is not the strongest quality control available. It is expensive, it takes a person away from fixing problems, and it produces a number with no statistical meaning: a defect rate of 0.4 percent across 1,500 pieces tells you less about the next lot than a structured sample does, because you learn nothing about how defects are distributed within the run.
ISO 2859-1 is a standard for sampling inspection by attributes. It assumes defects are scattered through a lot rather than clustered, and it trades a small amount of certainty for a large amount of time and money. You inspect a random sample, you count defective units, and the standard converts that count into a decision.
The phrase AQL, acceptable quality limit, names the defect rate you are willing to tolerate at a stated point in the distribution. It is not a target to aim for. A quality manager who says the team is "at 2.5" means the process is landing where the tolerance was set, not that anything improved.
For a lot of 26 to 50 pieces at AQL 2.5, the sample is 8 pieces and the acceptance number is 1. Pull eight at random, inspect them, count the defective units. Find one and the lot is accepted. Find two and it is rejected.
Accepted does not mean good. It means the sample did not provide enough evidence to refuse the lot. With eight pieces and a two-piece threshold, the arithmetic behind that decision assumes a real defect rate somewhere around two or three percent, so roughly one piece in forty arrives with a fault the inspection never looked for. That is the price of the method, and it is a lower price than the alternatives.
If one piece is at Ac and a second turns out to be defective, the lot fails. There is no averaging, no weighting by severity unless you weight by defect class first, and no second chance on the same sample. A fresh inspection needs a fresh sample and a fresh lot number.
General Inspection Level II is the normal default for finished goods. The two columns below are the numbers you will actually be quoting. Level I takes smaller samples and suits cases where every unit is expensive and the population is homogeneous. Level III roughly doubles the sample and belongs to destructive or life-critical testing.
A minor defect for a ceramic mug is a small glaze pinhole visible only when the piece is held to a light. A minor defect on a cotton tee is a stray fibre on the platen area that does not wash out. Both are recorded, neither is worth a remake, and both count against the AQL 4.0 column rather than the 2.5 column.
A failure is not an ending. It is a decision point, and the decision has a cost attached that depends entirely on which of four routes it takes.
Best when the ink is sound and the placement is wrong: registration off, graphic sitting 4 mm high, embroidery thread caught, a printing plate marked. Realigning and reprinting costs an hour of press time and no new setup. Screen setup runs 12,000 to 18,000 AMD per colour and DTF setup 8,000 to 15,000 AMD, so anything that avoids burning a new plate is worth checking first.
Needed when the fault cannot be corrected on the piece: a burnt plate, an ink batch that shifted hue, a colour outside tolerance against the approved reference. New screens are made, the run repeats, and the lead time moves. Add 3 to 5 working days to a standard 7 to 25 day schedule, which matters if the goods were timed for a trade show stand.
Legitimate for cosmetic faults on goods that still function: a mug printed on the reverse, an off-white blank, a notebook with a 2 mm trim variance. Sold at a discount or used for internal gifting. This route needs explicit written approval from the buyer and a separate line on the purchase order, because a discount nobody agreed to in advance becomes a dispute at invoicing.
Correct when the fault is structural: a chipped rim, a cracked handle, a food-contact surface that fails a scratch test, a wrong material or a wrong label. Also correct when a lot fails twice after rework. Rejected stock is scrapped or returned, and the lead time restarts from the point of failure, not from the original order date.
The honest answer is that it depends on which layer failed, and the reason buyers get unhappy is that this is usually never written down. Three layers produce merch: the blank supplier, our production, and the artwork. Each has its own error rate and its own cost.
A shade variation between two cartons of blanks is the supplier's cost. A print that ran 2 mm off registration or a plate that burnt mid-run is ours. Artwork supplied at the wrong size, with a missing bleed or a colour that cannot be matched to the available ink, is the buyer's. That third case is the one that generates the most friction, because the press did exactly what the file asked and the file was wrong.
Put the answer in the purchase order rather than in an email thread. Four clauses do most of the work: the AQL level and inspection scope, the defect classes and their thresholds, the remedy ladder with who funds each step, and the evidence required before a credit is issued. Then state the VAT treatment explicitly, because a credit note issued against a 20% VAT invoice has to adjust the tax line as well as the goods line, and doing that by hand invites a dispute six weeks later.
Claims are won or lost on documents, not on photographs of damage. Six items need to be in place before the truck leaves, and all six cost nothing to produce on the day.
A signed artwork proof showing the print at actual size on the actual blank, with the colour reference named on it. A packing list reconciled line by line against the purchase order, showing which units are in which carton. A photo set taken at packing time: a wide shot of the pallet, a close-up of each defect, and a coin or ruler in frame so the defect has scale. Without that reference object the close-up proves nothing about size.
Then a defect log listing quantity by class, a written disposition naming the route chosen and who authorised it, and a credit or remake note with the VAT breakdown in AMD. Timestamps matter as much as content. A photo taken at packing shows ink coverage; the same mug photographed three weeks later shows storage damage, and the two are not the same claim.
One more thing worth doing: photograph the cartons and the pallet wrap, not just the goods. Transit damage is far easier to attribute when the packing condition is recorded before the load leaves.
Defaulting to AQL 2.5 for everything is the most common mistake in this whole area, and it is wrong in both directions. It is too loose for goods that travel and too strict for goods that nobody looks at closely.
A ceramic mug fails in ways a cotton tee cannot. Glaze coverage varies across a curved surface, a rim chip makes the piece unusable and dangerous, and a printed mug that abrades under a dishwasher shows up as a defect that only appears after delivery. Drinkware usually justifies AQL 1.5 or 1.0 on major defects and a separate rule that any chipped rim is a critical finding regardless of count.
Knitwear is a different question. Seams twist, GSM varies between dye lots, and a size run will not match a nominal measurement exactly. Apparel tolerates AQL 2.5 on major and 4.0 on minor comfortably. Hard goods such as a USB drive or a power bank carry electrical risk, which is why those are usually inspected 100% for function rather than sampled for function at all.
When the stakes are high enough that a wrong accept is expensive, tightened inspection or double sampling is the right escalation. Double sampling lets you accept a good lot on the first small sample and sends you to a second larger sample when the first result is borderline, which suits repeat orders where you have prior data on the supplier.
Forty-five minutes for a lot up to 500 pieces, because drawing the sample and unpacking the cartons is most of the work. A 3,000-piece lot with several SKUs takes closer to two hours. It runs on the same day as packing, and we need the goods finished and boxed before it starts, not half-packed.
Either works, and they have different value. Your own person sees what you care about and gets a direct line to the workshop floor. A third party works to a written specification you supply in advance, which is what makes their report usable later as evidence. What we cannot offer is an accredited inspection body; see the next question but one.
Yes, and any supplier who tells you otherwise is overpromising. Sampling controls the risk you accept, it does not remove it. A pass at AQL 2.5 on a sample of 50 still means a small residual rate of unseen faults. What sampling does guarantee is that a defect found in the sample was found before 1,500 pieces shipped rather than after.
No. It is a method for drawing a sample and turning the result into a decision, not a certification scheme and not a certificate anyone can issue. Using the standard correctly means stating the inspection level, the AQL, the lot size and the date on the inspection record, and that record is the document that matters.
No. Corp-Merch is an independent merch producer. We are not an accredited inspection company, we hold no certificate that qualifies us to issue ISO certificates, and nothing on this page should be read as an endorsement by any standards body, certification scheme or inspection agency. Every number here describes how we run our own production.
Three to five working days added to the original schedule for a screen reprint, since the plates have to be burned again. A DTG reprint is faster, usually two days, which is why it is our recommendation for small batches where a second setup charge is hard to justify. A full rejection restarts the whole run from the failure point.
A chipped or cracked rim, a cracked handle, and any surface on the food-contact area that scratches under a fingernail test. These are critical because the piece cannot safely be used or sold, and one is enough to fail a lot regardless of the AQL. A glaze pinhole on the side of the body is minor.
No, and asking for one is usually a false economy at that size. Below about 100 pieces, inspecting every item costs less in time than drawing a random sample and reporting on it. Our minimum for screen and DTF is 50 pieces, for sublimation 25 to 50, and for DTG as low as 1, so these runs are small enough to check by hand. Save the statistical method for orders where the maths earns its keep.
Corp-Merch is an independent corporate merch producer based in Yerevan, with a production unit in Vanadzor. We are not affiliated with, sponsored by or endorsed by any certification body, inspection agency or standards organisation. References to ISO 2859-1 describe a published sampling method we apply to our own work. Named companies such as AmeriBank, Ardshinbank, Converse Bank and HayPost appear only as examples of the kind of buyer we produce for. All figures in AMD are quoted before 20% VAT.