Most apparel buyers inspect every lot the same way. ISO 2859-1 says you shouldn't.
How the clause 9.3 switching rules work — and what staying on Normal actually costs.
The half of the standard almost nobody uses
Ask any QA manager which AQL they inspect at and you'll get a fast answer: 2.5 major, 4.0 minor. Ask them when they last moved a factory to Tightened inspection, or claimed Reduced on a factory that hasn't failed a lot in a year, and the room goes quiet.
That's the problem. ISO 2859-1 is not a table of sample sizes. It's a sampling system — sample sizes, acceptance numbers, and a set of rules that move a factory between three levels of severity based on its own recent history. Clause 9.3. Without the switching rules, you are using a fraction of the standard, and the standard itself says so: ISO 2859-1 plans are designed to be used with switching in a continuing series of lots.
The three severities
Normal — the default. Table 2-A. Where every factory starts.
Tightened — Table 2-B. Same sample size, lower acceptance number. A lot that scraped through on Normal fails on Tightened. This is what a deteriorating supplier is supposed to feel.
Reduced — Table 2-C. A smaller sample for a factory that has earned it. This is the commercial payoff of a good supplier relationship, and it's the one buyers almost never claim.
The rules, in plain English
- Normal → Tightened: two rejected lots out of the last five.
- Tightened → Normal: five consecutive accepted lots.
- Tightened → discontinue: if tightened inspection isn't cleared within five lots, stop accepting from that source until the supplier fixes something. This is the clause with teeth, and the one most often ignored.
- Normal → Reduced: a switching score of 30 with steady production — and only where the responsible authority permits it. Reduced is never automatic.
- Reduced → Normal: one rejected lot, or production that is no longer steady.
Note what the rules are indexed to: the factory's own recent record, per buyer, per AQL pair. Not a gut call, not a relationship, not "they've been fine lately."
Why staying on Normal is a decision, not a default
A static Normal plan applied to a factory that is quietly drifting does exactly what the math says it will: it accepts lots at roughly 95 % when they sit at the AQL, and it keeps doing that while quality slides underneath. The escapes you find at the DC were not a failure of AQL sampling. They were a failure to switch.
The mirror image is just as expensive in the other direction. A factory with twenty clean lots behind it is still drawing full Normal samples — inspector hours, factory hours, and calendar days you are paying for and no longer need. Reduced exists precisely to give that back, and the fee is sitting there unclaimed.
The reason it doesn't happen
Switching is bookkeeping. To know a factory's severity today you need the outcome of its last five lots, the running switching score, and a judgement on whether production is steady — per factory, per buyer, per AQL pair. On a spreadsheet, across thirty factories, that's a job nobody has time for. So everyone stays on Normal, forever, and calls it consistency.
How QualityIris handles it
QualityIris implements the clause 9.3 switching rules as an advisory layer. Before the inspector opens a lot, the app already knows that factory's standing — normal, tightened or reduced — from its own inspection history, tracked per factory, per buyer, per AQL pair, and shows it as a severity recommendation.
Three things we were deliberate about:
- The inspector stays in charge. The recommendation can be overridden, but the override requires a written reason, and that reason prints on the buyer-facing audit PDF next to the severity actually used. The judgement is human; the audit trail is not optional.
- Reduced is permitted, never automatic. The app can tell you the score has reached 30 with steady production. It won't quietly shrink your sample. That call belongs to the responsible authority, exactly as the standard says.
- The numbers come from the right table. Acceptance numbers resolve from ISO 2859-1:1999 Table 2-A (normal), 2-B (tightened) and 2-C (reduced), by sample size code letter, with the clause 10.3 arrows moving the code letter — not just the acceptance number.
One thing that is not there, on purpose: ISO 2859-2 (LQ) audits have no switching. An isolated lot has no continuing series to switch against, so we hide it in LQ mode rather than invent a severity that the standard doesn't define.
Where to start, even without software
You don't need us to begin. Pick your three highest-volume factories. Pull the accept/reject outcome of their last five lots. Any of them at two rejects out of five should be on Tightened this week, and any of them with a long clean run is a Reduced conversation with your responsible authority. That single exercise usually pays for itself before the month is out.
If you want the arithmetic done for you — what the escapes cost, and what the Reduced fee you're not claiming is worth — the AQL Reality Check calculator is free and needs no sign-up.
More on how switching is implemented: https://qualityiris.com/lp/iso-2859-1-switching-rules
