Last March a lot of 240,000 crimp-top 2 mL vials with caps failed outgoing inspection at a customer's dock, and the reason was not the glass. The receiving lab pulled 125 pieces per ISO 2859-1 at AQL 0.65 for major defects, found four vials with septa pushed proud of the crimp skirt by roughly 0.4 mm, and rejected the shipment. My own outgoing record for that lot had passed at AQL 1.0 with a sample of 80. Two sampling plans, two answers, one lot. That gap is what this article is about: how to set an AQL sampling plan and build an outgoing QC record for vials with caps that holds up when the buyer runs their own plan against yours.
Most failures I see in outgoing QC for vials with caps are not manufacturing failures. They are plan failures. The sample size was too small to catch a 0.3% defect rate, or the defect class was wrong, or the record did not say which plan was used, so nobody could reconcile two different results. Getting the AQL sampling plan right is the whole game.
I see a lot of suppliers default to AQL 1.5 across the board because that is what their customer's purchase order says. That number is a starting point, not a decision. The AQL should come from what happens downstream when a defective unit reaches the analytical bench.
Here is how I classify defects for vials with caps, and the AQL I assign to each class. This is the table I hand to a new QC tech on day one.
| Defect class | Example defects | Typical AQL | Inspection level |
|---|---|---|---|
| Critical | Cap not seated (loose in bag), wrong cap size on vial, cracked neck, septum missing, glass shard in cavity | 0.065 (zero accept on sample) | II, tightened |
| Major A | Septum proud of crimp skirt, cap cocked more than 0.3 mm, visible chip at sealing surface, incomplete crimp (skirt not folded) | 0.65 | II |
| Major B | Scratches on body, cosmetic cap discoloration, label misregistration, minor dimensional drift within tolerance | 1.5 | II |
| Minor | Cosmetic blemishes that do not affect function, minor mold marks on cap | 4.0 | II |
The critical class gets zero accept. If you find one critical defect in the sample, the lot is rejected and you sort 100% or scrap it. That is not negotiable for a container that will hold a sample going into an autosampler. A cap that is loose in the bag will fall off in the tray and the autosampler will fault, or worse, the needle will punch air.
The AQL 0.65 for major defects is the number I have converged on after years of arguing with buyers. It is tight enough to catch the septa-proud problem at a realistic rate, and loose enough that a good lot does not get rejected on a single borderline piece. If your customer insists on 0.4, fine, but understand you are now sampling harder and rejecting more lots that are functionally acceptable. That cost lands somewhere.
The single most common error I see in outgoing inspection records is a sample size that does not match the lot size and inspection level. People pull 32 pieces because that is what they pulled last time. That is not a plan, that is a habit.
ISO 2859-1 (the same table structure appears in ANSI/ASQ Z1.4) gives you the sample size code letter from lot size and inspection level. General inspection level II is the default. Here is the working table for the lot sizes we actually ship, at AQL 0.65:
| Lot size | Code letter (level II) | Sample size | Accept / reject at AQL 0.65 |
|---|---|---|---|
| 501 to 1,200 | H | 50 | 1 / 2 |
| 1,201 to 3,200 | J | 80 | 2 / 3 |
| 3,201 to 10,000 | K | 125 | 3 / 4 |
| 10,001 to 35,000 | L | 200 | 5 / 6 |
| 35,001 to 150,000 | M | 315 | 7 / 8 |
| 150,001 to 500,000 | N | 500 | 10 / 11 |
Read the accept/reject column carefully. At lot size 3,201 to 10,000, sample 125, AQL 0.65, you accept on 3 defects and reject on 4. That is the plan that would have caught my 240,000-piece lot. The customer's 125-piece sample was correct for their lot size at level II. My 80-piece sample was correct for a smaller lot size. The plan was right; the lot size assumption was wrong because I had split the production run across two shifts and sampled them as one lot.
That is a real failure mode. If you produce a lot in two shifts on two machines, either treat them as two lots and sample each, or treat them as one lot and sample as one. Do not sample as if it were one lot and then report it as two. The record has to match the physical reality.
ISO 2859-1 has a switching rule set that most suppliers ignore. If you have had two out of five consecutive lots rejected on original inspection, you switch to tightened inspection. If you have had ten consecutive lots accepted on original inspection, you are eligible for reduced inspection. In practice, most suppliers run normal inspection forever and never switch. That is a mistake for two reasons.
Tightened inspection after a bad run is how you prove to a buyer that you caught the problem and fixed it. It is a written signal that the process is under scrutiny. Without it, the buyer sees a rejection followed by a pass with no explanation.
Reduced inspection is where you get your inspection cost back. After ten lots of clean history, reduced inspection drops your sample size and lets you run a leaner plan. I have seen a supplier cut outgoing inspection labor by 40% on stable product lines by actually running the switching rules and documenting them. The buyer accepted it because the rule set is in the standard, not invented by the supplier.
Document the switch date and the trigger in the outgoing record. "Switched to tightened on 2024-03-14 after two rejections in lots 24-0188 and 24-0201" is the kind of line that survives an audit.
I have audited maybe thirty supplier QC records for vials with caps. The good ones share a structure. The bad ones are a signed sheet that says "PASS" and nothing else.
Here is what I require on every outgoing record, and what I put on ours:
The measurement data is the line most suppliers skip. A pass/fail on crimp diameter tells the buyer nothing about whether the lot is drifting toward the limit. If the mean crimp diameter is 8.62 mm against a spec of 8.60 ± 0.10 mm, that is a lot that will pass today and fail in three months. The buyer cannot see that from a checkmark. I put the mean and the range on the record because it is the early warning that keeps a lot from becoming a rejection.
If you are sourcing vials with caps and your supplier cannot produce a record with these fields, that is a finding. See Vials With Caps for the product side of the same conversation, and Vial Supplier for how to qualify a source on documentation rather than price.
The 240,000-piece lot I opened with is worth walking through because it shows how a defect class and a sample size interact.
Symptom: the receiving lab rejected four vials out of 125 with septa proud of the crimp skirt. The septa sat roughly 0.4 mm above the top of the aluminum skirt instead of flush or slightly recessed.
Measurement that identified it: a go/no-go fixture with a 0.2 mm step. Any septum more than 0.2 mm proud fails. We had been checking with a flat plate and a visual, which caught gross failures but not 0.4 mm. The fixture change was the fix on the measurement side.
Root cause: the crimping head on line 2 had drifted on its depth setting by about 0.15 mm over the run, which was enough to leave some caps under-crimped. Not enough to fail a pull test, but enough to leave the septum proud.
Fix: we added the go/no-go fixture to the outgoing inspection for major A defects, tightened the AQL for that class from 1.0 to 0.65, and added a crimp-depth check every 500 pieces during the run. The next three lots passed at AQL 0.65 with zero major A findings.
The lesson is not that the crimper drifted. Crimpers drift. The lesson is that the defect class was assigned an AQL that was too loose for the consequence, and the measurement method was too coarse to catch the defect at the rate it was occurring. Both had to change.
Symptom: a customer running 9 mm screw-thread vials reported that about one in eight vials would not thread into their autosampler tray without cross-threading. They were about to reject the lot.
Measurement: we pulled 80 pieces at random from the retained sample and ran a thread gauge. Sixty-eight threaded cleanly, twelve bound at the first turn. That is a 15% defect rate, which is enormous. But the original outgoing sample was 32 pieces, and the defect rate in the production population was actually about 0.4% because the twelve bad pieces were all from one cavity on the injection mold. The sample of 32 had not hit that cavity.
Root cause: one of the eight cavities on the cap mold had a worn thread insert. Caps from that cavity were out of tolerance on the thread major diameter by 0.08 mm, enough to bind on a 9 mm screw thread with a 0.05 mm clearance.
Fix: we changed the sampling to a stratified pull across cavities, not a random pull from the blended bin. For a cap mold with eight cavities, we pull at least 10 pieces per cavity in the outgoing sample. That is 80 pieces minimum, which lines up with the ISO 2859-1 sample size for a lot in the 1,201 to 3,200 range at level II. The cavity-stratified pull is what catches a single-cavity defect at a rate a blended random sample will miss.
Any supplier running multi-cavity molds for caps or multi-cavity glass forming should be stratifying the outgoing sample by cavity. Ask for it. If they cannot tell you which cavity a defect came from, they are not stratifying.
Outgoing inspection for vials with caps is not just a visual for chips and scratches. There are four dimensional checks that catch the defects that actually cause problems downstream. Here they are with the numbers I use for a standard 2 mL crimp vial with an 11 mm aluminum cap:
For screw-thread vials, add a thread gauge check and a cap torque check. A 9 mm screw cap should seat at 0.6 to 0.9 N·m for a hand-tightened cap on a glass vial. Below 0.5 N·m the cap will back off in transport. Above 1.2 N·m you risk cracking the neck on a thin-wall vial. I have seen a lot rejected because the capping machine was running at 1.4 N·m and cracking one in 400 necks. The outgoing torque check caught it.
The dimensional checks and the visual checks go in the same record. The sample size is the same sample. You are not pulling a separate sample for dimensions. That is the mistake that doubles inspection labor for no gain.
A static leak test at outgoing inspection proves the cap is seated and the septum is intact. It does not prove the container will hold vacuum through a 20-minute autoclave cycle or a freeze-thaw. I have written elsewhere about the four leak paths and only one showing up in a static test, so I will not repeat it here. The point for outgoing QC is that the leak test is a screen, not a qualification. Run it at 100% or at the sample size, but do not let a passing leak test substitute for the dimensional checks. A vial with a 0.4 mm proud septum can pass a static leak test and still fail in the autosampler.
For labs running the receiving side of this, the same logic applies to the analytical method. The Lab Chromatography side of the house cares about injection reproducibility and carryover, which is downstream of whether the cap sealed. A cap that seals but sits proud will change the injection depth and the RSD. I have seen RSD% jump from 0.4% to 1.8% on a 10-injection sequence because the septa were sitting unevenly. The fix was a cap-seating check, not a method change.
When a lot fails outgoing inspection, or when a buyer rejects a lot you passed, work through this in order. The first three items catch most of the real problems.
If you get through all seven and the lot still fails, the problem is upstream in the process, not in the sampling plan. That is a different conversation, and it belongs to the in-process capability discussion, not to outgoing QC.
An AQL sampling plan is only as good as the contract that references it. If your purchase order says "vials with caps, 2 mL, 11 mm crimp" and nothing else, you have no basis to reject a lot on sampling grounds. Put the plan in the PO.
I specify four things on every PO for vials with caps:
That is a one-page appendix. It is the difference between a plan you can enforce and a plan you hope the supplier is running. For the analytical side of the same supply chain, the chromatography consumables and the HPLC filter specs go in the same appendix, because a vial that passes outgoing QC and then fails in the autosampler is a system problem, not a vial problem.
The most common audit finding on outgoing inspection records is not the sampling plan. It is the gauge calibration. The go/no-go fixture, the crimp diameter micrometer, the torque wrench, the thread gauge: every one of them needs a calibration record with a due date, and the record needs to show the calibration was current on the inspection date.
I have seen a supplier fail an audit because the micrometer used on the outgoing record had a calibration due date two weeks before the inspection. The measurement was probably fine. The record was not. That is a finding.
Put the gauge ID and calibration due date on every outgoing record. It is two fields. It saves the audit. The Aijiren chromatography consumables and Thermo Fisher sample handling documentation follows the same principle: the measurement is only as good as the instrument that produced it, and the instrument record travels with the result.
Two plans will disagree on the same lot. That is expected. ISO 2859-1 is a sampling scheme, not a measurement of the lot. A lot at a true 0.4% defect rate will pass a 125-piece sample at AQL 0.65 about 85% of the time and fail about 15% of the time. That is the nature of sampling.
The way to handle a disagreement is to go to the retained sample and the measurement data, not to argue about the plan. If the buyer found four defects in 125 pieces, you pull your retained 20 pieces and check them with the same fixture. If your retained sample is clean, you have a case that the defect is not uniform across the lot. If it is not clean, you have a process problem and you sort the lot.
The suppliers who handle this well are the ones who keep the retained sample and the measurement data. The ones who handle it badly are the ones who say "our plan passed" and stop there. The plan is not the defense. The data is.
For the regulatory frame on container and sampling requirements, the U.S. EPA SW-846 test methods and the method-specific container requirements are the reference most environmental labs work to. The Restek technical resources cover the analytical side of the same sample handling chain, and they are worth reading alongside the container spec because the two have to match.
Set the AQL from the consequence of the defect, not from the PO default: critical at 0.065 with zero accept, major A at 0.65, major B at 1.5, minor at 4.0. Pull the sample size from ISO 2859-1 at general inspection level II for the actual lot size, stratify across cavities and shifts, and run the switching rules so tightened inspection follows a rejection and reduced inspection follows ten clean lots. Record the measured values, the gauge calibration, and the retained sample location on every outgoing record, because the plan is a screen and the data is the defense when a buyer's plan disagrees with yours.
Related resources from our lab consumables network:
External references and standards: