Home › Field notes › ISO 2859-1 AQL Sampling for Bags: Levels, Plan Tables and Defect Class

Modular Bag Pro · 2026-10-11 · 3,685 words · 17 min read

ISO 2859-1 AQL sampling for bags decides exactly one thing: whether the lot presented may move forward, based on counted nonconforming units in a sample drawn under a declared inspection level. The parameters that make the verdict defensible are lot identity and size, the general inspection level selected, and the AQL value attached to each defect class, with our published baseline sitting at level II and AQL 2.5 with Critical 0, Major 2.5 and Minor 4.0 against an order minimum of 500 pieces. Everything else that buyers care about sits outside this arithmetic, because a count taken at the end of the run cannot certify the units never opened and cannot examine construction sealed inside the assembly. Read the plan as release-or-hold evidence for one batch, and carry questions of durability, fit and hidden workmanship on first-piece records, in-process patrol logs and pre-production validation instead.

What ISO 2859-1 AQL Sampling Actually Decides About a Bag Lot

An attribute sampling plan answers one narrow question: given the evidence counted in this sample, may this lot go forward or must it be held. Everything else that a buyer cares about — how long the seam lasts, whether the harness fits the population it was drawn for, whether a coating survives a season — sits outside that verdict. Treating a passed inspection as proof of durability is the most expensive misunderstanding in this subject, because it quietly transfers risk from evidence to assumption.

The system published by ISO 2859-1 works with counted units rather than measured values. Each sampled bag is judged conforming or nonconforming against written rules; the counts are compared with acceptance and rejection numbers taken from a plan table; the lot is then released or held on arithmetic alone. Nothing is averaged, no tolerance band is inferred, and no single measurement is projected across the shipment.

Three inputs have to be fixed before a carton opens, and each one belongs in the purchase document rather than in an email thread. The first is how a lot is defined — one order line, one colourway, one output batch, or some other agreed boundary. The second is the inspection level, which together with lot size returns the sample size code letter. The third is the AQL value attached to each defect class, each carrying its own accept and reject pair. Altering any one of them after counting begins invalidates the exercise.

A useful way to hold the boundary is to separate three questions that teams habitually collapse into one. Does this batch match the approved reference and the defect library? That is what the plan decides. Will the construction survive its service life? That is answered by validation on a pre-production build. Does what was sewn match what was drawn? That is answered by first-piece evidence and by records kept while the line runs.

Selection rule: Use ISO 2859-1 to decide the fate of the batch in front of you, and answer durability, fit and hidden workmanship through first-piece photographs, in-process patrol logs and validation testing, because counting released units says nothing about any of them.

General Inspection Levels and Special Levels: How Much Discrimination You Buy

Inspection level is the dial that sets statistical sensitivity. Three general levels sit on that dial, the middle one acting as the reference position in most consumer-goods programmes because it balances the labour of opening cartons against the chance of detecting a deteriorated batch. Stepping up buys a larger sample and a sharper ability to separate a good lot from a marginal one; stepping down saves inspection hours and accepts a blunter instrument.

Special levels occupy a different corner entirely. They exist for checks that are destructive, expensive, slow or physically awkward to run across many pieces — a laboratory measurement rather than a visual look. The samples they return are small, so they carry much weaker protection, and no programme should adopt one for appearance or function work merely to save a day. A destructive teardown legitimately belongs here, because cutting a seam open consumes the unit.

Conditions specific to sewn goods should drive the choice rather than habit. Output produced in one colourway on one line during one short window behaves more homogeneously than the same quantity consolidated from two lines across three shifts, and a homogeneous population needs less discrimination to reach the same confidence. Carton presentation matters too, since opening and re-packing hundreds of boxes consumes hours and risks the very damage later attributed to transit. Where a programme already carries a marginal history, moving one general level up usually costs less than the write-down triggered by one missed lot.

Whatever is chosen has to be written down and then repeated on every subsequent shipment of that programme. Ad hoc switching — a bigger sample this time because somebody complained — removes the comparability that gives the record value, and it quietly destroys the basis for the switching rules that later decide whether discipline can relax or must tighten.

Special levels deserve a separate line of reasoning, because they solve a different problem. Their purpose is to keep a destructive or costly check affordable, not to make a routine appearance check faster. A team that tears down a handful of units to confirm construction nobody can see later is using them correctly; a team that looks at six pieces because the truck is waiting is not.

Verdict: Fix the inspection level once, write the reason for that level into the purchase document, and keep it identical across every shipment of the programme, because changing it between batches destroys both comparability and the standing of the switching rules.

Reading a Sampling Plan Row Without Inventing a Threshold

Every row in a plan table carries the same four-part structure, and understanding that structure is enough to read any published version correctly. Lot size combined with inspection level returns a code letter. The code letter points to a sample size. The chosen AQL selects a column. The intersection yields an acceptance number and a rejection number for that class, stated as counts of nonconforming units found inside the sample.

The arithmetic is then unremarkable. Count the offending units found for the class. If the count sits at or below the acceptance number, that channel passes. Once it reaches the rejection number, the lot cannot go forward. A worked but purely illustrative example: imagine the plan returns an acceptance number of five and a rejection number of six for majors, and the inspector finds six — the shipment is held regardless of how trivial the sixth unit looks, because these numbers are counts rather than opinions.

Plans are published in single, double and multiple forms. A single form takes one sample and decides. A double form takes a first sample, decides when the outcome is comfortably good or comfortably bad, and draws a second sample when the result lands between those two points. Multiple forms extend the logic further. The families are arranged so protection behaves comparably across them; the real differences are administrative burden and how many units end up inspected on average, not licence to redraw until the answer changes.

What belongs on the record is broader than the verdict itself. Retain lot identity and size, inspection level, plan form, the AQL figures per class, counts found per class, the accept and reject numbers applied, who inspected and when, and the revision reference of the defect library in force. Those fields are what convert a release note into evidence that still stands when a retail partner asks, months later, why a particular shipment went out.

Judgement: Print lot identity, inspection level, plan form, class-wise AQL values, counts found and the defect library revision on every report, because a verdict recorded without those fields cannot be reconstructed or defended once cartons have left.

Critical, Major and Minor for Bags: A Library an Inspector Can Apply

Defect classes are not severity adjectives; they are decision channels. Each channel carries its own AQL value, its own accept and reject pair, and therefore its own consequence. Filing a finding in the wrong channel is the most common origin of post-shipment argument, because the identical defect can pass the batch under one heading and fail it under the next.

Critical covers what makes the article unsafe, unlawful or unusable for its declared purpose: a needle fragment lodged inside filling, a sharp edge left on hardware, a closure that traps skin, a mandatory marking omitted, or a compliance breach in a market that demands the document. Because the consequence is not negotiable, a critical channel runs at zero acceptance — a single occurrence holds the lot. It should never be used to win an argument about appearance.

Major covers functional failures a customer notices and returns for: a seam that has opened, a slider that refuses to run, a webbing anchor whose reinforcement was skipped, printed decoration releasing from its film, attachment rows that cannot be threaded because repeat drifted, or a module that is not the one on the approved list. Minor covers workmanship deviation inside tolerance that a customer may see but is unlikely to return for — slight shade difference between two panels, uneven webbing tails, or cosmetic stitch irregularity along a non-structural run.

The test of a usable library is whether two inspectors reach the same verdict on the same unit. That requires definitions written as observable conditions, backed by photographs and, wherever relevant, a stated measurement method with tolerance. Words such as "poor workmanship" or "unacceptable appearance" fail this test every time. Where a defect sits close to a boundary, the library should already say which side it falls on, because the person holding the unit has seconds to decide and should not be interpreting intent.

Bottom line: Write each defect as an observable condition with a photograph, a tolerance where relevant and a fixed channel, so that Major 2.5 and Minor 4.0 mean the same thing to the inspector, to us and to the brand.

Why AQL 2.5 Is a Starting Position Rather Than a Verdict

The letters abbreviate acceptance quality limit, and the phrase is habitually misread as a promised defect rate. It is not. It describes the quality level the scheme is constructed around: populations at that level are the ones the plan is arranged to accept most of the time, while materially worse populations face a rising probability of rejection. A shipment may therefore be released at AQL 2.5 and still contain defective units, because sampling inspects a subset and both parties accept that risk.

The baseline we publish — general inspection level II overall at AQL 2.5, with Critical 0, Major 2.5 and Minor 4.0 — is a commercial starting position that must be written into the order before any carton opens, not a technical conclusion. A brand selling through a channel with expensive return handling may want the major figure tightened. A promotional programme where nobody inspects behind an inner liner may accept a looser minor figure. Both positions are legitimate, and both must be agreed on paper rather than discovered during a disagreement.

One property regularly surprises buyers: moving an AQL value does not always move the plan. At small sample sizes several different AQL values share the same accept and reject pair, so a headline change can return identical instructions at the bench. Negotiation should therefore focus on what the table actually produces — sample size together with the accept and reject numbers — instead of on the figure printed in a contract.

The zero-acceptance position held for criticals deserves its own note. It is not statistical decoration; it reflects that some findings cannot be tolerated in any count at all. When one surfaces, despatch stops, the unit and its immediate neighbours go into quarantine, and the corrective route addresses the operation or component that produced it. Counting resumes only once the cause is contained.

Takeaway: Agree the AQL figures in writing, then inspect the plan those figures actually return — sample size plus the accept and reject pair per class — because it is the returned pair that governs the despatch, not the headline number.

When a Lot Fails: Counts, Switching Rules and Containment

A failed channel fails the batch. Objections along the lines of "it was only one above the line" misunderstand what acceptance numbers are for, since they already carry whatever tolerance the two parties agreed. The response should be administrative rather than argumentative: quarantine the shipment, stop related despatch, and notify whoever owns the order with the recorded counts and photographs attached.

The system also contains switching rules that move discipline across successive batches. A programme whose lots release consistently may qualify for a reduced burden; a programme meeting the trigger conditions set out in the cited version moves to tightened inspection, with suspension available where failure repeats. Those triggers belong to the method version named in the order, and the reason for keeping unbroken sequential records is precisely that the switching position can then be established without an argument.

Containment has a technical side as well. Whatever surfaced at the end was present earlier in the run, so the live question is which operation, material batch, machine setting or shift produced it, and how wide the affected population is. Sometimes that reaches goods already despatched, sometimes stock still on the line, and sometimes an incoming component. Cases like an attachment grid drifting out of row pitch are instructive, because the failure belongs to cutting or sewing rather than to anything a counter could have found reliably at the end.

Escalation also runs against a commercial clock. Where replacement stock is needed, air transit occupies 5-8 days, courier 3-5 days, and sea freight 25-35 days, so the decision to hold the conforming remainder or move it has to be taken with those windows and the retail calendar in view rather than after they have closed.

Spec rule: Record the disputed outcome as counts against the printed plan, then move the commercial discussion to recovery windows of 5-8 days by air, 3-5 days by courier or 25-35 days by sea, because elapsed days decide what can still reach the season.

Re-inspection and Rectification: What a Second Look Can Repair

A held batch normally has two routes forward: accept it under a concession negotiated with the brand, or rectify and resubmit. Rectification nearly always means sorting — examining every unit against the library and repairing or replacing failures. The method works and it is honest, provided everyone accepts that a sort removes only what the list describes. An instruction reading "check the stitching" will never surface a dimensional drift the list omitted.

Resubmission is a new event, not a continuation of the old one. The standing switching position and any specially agreed procedure govern how the rectified batch is examined, and plenty of programmes deliberately look at rectified stock under a tighter setting because the population has already demonstrated elevated risk. Treating re-inspection as a free second draw from the same plan is the error that returns as complaints three shipments later.

The cause deserves more attention than the repair. In practice a concentrated cluster of one defect maps onto something specific: a needle change that altered stitch behaviour, a cutting die that drifted, a substitute batch of webbing, an operator newly placed on one station, or a machine whose setting was adjusted mid-run. Correcting that cause costs far less than sorting twice, and it is the difference between an incident and a pattern. Buyers running development programmes for modular platforms generally see the same handful of causes repeating, which is why component approval before bulk cutting repays itself on every subsequent order.

Cost and schedule follow the decision. Sorting consumes labour and calendar days, packing materials are consumed twice, and the freight plan has to be rebuilt around whatever window remains. Teams that dislike this outcome usually spend slightly earlier instead — component approval, photographic review of a first piece, patrol inspection during the run — because at those stages the defect is still cheap to fix.

A reasonable filing discipline keeps the whole matter small: the original report, the cause statement, the corrective action, the sort instruction actually issued, the resubmission report and the retained reference units stored together under one order number. Without that bundle, the next disagreement starts from nothing.

Where Acceptance Sampling Stops: The Layer Counting Cannot Replace

Final inspection sits at the end of a chain and can examine only what is visible in the units placed in front of the inspector. Anything sealed inside the assembly — backing material behind an attachment face, reinforcement buried under a bound edge, allowance inside a turned join, foam under a back panel — is either already right or already wrong, and no volume of counting will reveal which. Those belong on the first-piece review, in a cut sample, and in photographs kept against the order.

Material conformance needs its own route and predates any finished-goods verdict. Cloth identity and tensile behaviour are qualified by component methods such as ASTM D5034, resistance to surface wear through rotary abrasion such as ASTM D3884 or the Martindale route in the ISO 12947 family, and surface water behaviour of flat cloth through AATCC 127. Each qualifies a material rather than a carrier: seams, needle holes, slider paths and attachment stitch lines determine real exposure, and they can only be judged on the assembled article.

Distribution and market access sit outside the plan too. Parcel handling evidence follows distribution simulation built on ISTA methods; restricted substances follow declarations under REACH (EC 1907/2006) and, where the destination or customer list requires, an exposure review against California Prop 65, CPSIA documents where user scope brings them into play, and OEKO-TEX Standard 100 inside its certificate scope. Counting defects proves none of these.

The practical conclusion is a layered file: components approved before cutting, a reviewed and photographed first piece, patrol records taken during the run, the lot verdict at the end, and a reference unit kept in store. On that last point, the SGS-verified production base we work with holds a 4,950 m² floor running 149 machines and 7 production lines with 137 people, planned around 200,000 units each month; behind it sits QUANZHOU JUNYUAN BAGS, established in 2014, whose founder has worked in bag production since 2004. Orders move through the same gates whatever the construction — 6-10 working days of sampling, 12-15 where the build is complex, a pre-production reference for sign-off, mass production over 35-50 days — and close with the reported result at AQL 2.5 before anything is loaded.

Writing the Plan into the Order So Disputes Are Settled by Paper

A sampling clause needs surprisingly little text to do its job. It should name the cited version of the standard, the inspection level, the plan form, the lot definition, the AQL value and accept and reject pair per class, the revision of the defect library, who performs the count, whether the count is witnessed, what happens after a hold, and who bears the cost of sorting. With those eight items written down, most disagreements resolve by reference rather than by negotiation.

Commercial terms belong in the same document, so nobody has to reconstruct them under pressure. The order minimum is 500 pieces, quoted prices carry FOB Xiamen terms with T/T 30/70 settlement, sampling runs 6-10 working days and stretches to 12-15 when the interface is complicated, and mass production occupies 35-50 days once approvals and inputs have closed. Tooling and screens sit in a USD 300-2,500 band and sampling carries a USD 50-150 fee that is refunded against the order, which makes the cost of doing the evidence properly easier to accept early.

Timelines then line up in one sequence: freeze the drawing and library during sampling, review the first piece before cutting, hold the count at the end, then choose the freight mode against the selling calendar. A programme that skips the middle steps does not save time; it relocates the same delay to a point where air freight at 5-8 days is the only way to recover it.

Buyers who want this handled as one documented route can start from the specification and inspection services description and adapt it, or ask for the standard clause we use when the brand has no template of its own. Either way the discipline is identical: decide the rules before the cartons arrive, then apply them without exception, because a plan applied selectively is worth less than no plan at all.