AQL 1.0 on major defects is the wrong default for a 12 x 32 mm vial--Aijiren Sample Vials

A 500-vial shipment of 9 mm screw-thread vials failed the receiving lab’s incoming check on three consecutive lots, and the supplier’s certificate of analysis showed zero defects on all three. The customer pulled 32 vials per lot, found two with a 0.4 mm short fill and one with a septum seated 0.6 mm proud of the cap skirt, and asked a simple question: what sampling plan produced that “zero”? The answer was none. The supplier was doing a 10-piece visual pull at end of shift. That is not an AQL sampling plan, and it is not outgoing QC. It is a habit.

If you buy oem vial product at volume, the AQL sampling plan and outgoing QC record are the two documents that decide whether a nonconformance is your problem or the supplier’s. This article is about how to specify both, what the numbers mean, and what a real inspection record looks like when it is done properly.

Most suppliers default to AQL 2.5 on major and AQL 4.0 on minor, because that is what their generic inspection sheet says. For glass vials destined for HPLC autosamplers, those numbers are too loose. An AQL of 2.5 on a major defect means you are contractually accepting up to 2.5% of the lot with that defect. On a 10,000-piece lot, that is 250 vials with a dimensional or closure defect, and the autosampler does not care that you agreed to it.

I specify AQL 0.65 on critical, AQL 1.0 on major, and AQL 2.5 on minor for oem vial shipments. That is not arbitrary. It maps to what a 100-position autosampler tray can absorb before a run is compromised. If one vial in 150 has a bent skirt that causes a needle strike or a cap that will not seat, you will see it inside a single sequence. AQL 1.0 on major gives you a reject decision at the lot level before that vial reaches the tray.

The AQL level is not the same as the defect rate you will see. It is the quality level that has a defined probability of acceptance (usually 95%) under the sampling plan. A lot at exactly AQL 1.0 will still be rejected roughly 5% of the time by design. That is the point. The plan is a gate, not a guarantee.

Critical, major, minor: the classification that actually matters

Defect classification is where most supplier QC sheets fall apart. They list “dimension” as one line item. That is useless. Here is the classification I write into every purchase specification for a 12 x 32 mm crimp or screw vial:

Critical (AQL 0.65, zero accept on the sample if any found): glass cracks, chips at the finish, incomplete threads on screw vials, septum not bonded or missing, cap not engaging threads, foreign particulate inside the vial, fill volume short by more than 5%.
Major (AQL 1.0): body OD outside 12.00 ± 0.20 mm, height outside 32.00 ± 0.50 mm, neck ID outside tolerance, septum thickness outside 1.0 ± 0.1 mm, cap torque below 0.4 N·m or above 0.9 N·m on a screw closure, crimp skirt not fully rolled, visible glass seam or mold line in the seal zone.
Minor (AQL 2.5): label misalignment, cosmetic scratches outside the seal zone, print legibility on the cap, minor color variation in the septa batch.

Note that critical defects get a zero-accept rule on the sample even when the AQL math would allow one. That is common practice and I recommend it for glass and closure integrity. One cracked vial in a sample of 200 is a process signal, not a sampling artifact.

ISO 2859-1 general inspection level II is the sample size you should be asking for

The sample size question has a correct answer and it is not “whatever the supplier sends.” Use ISO 2859-1 (the international twin of ANSI/ASQ Z1.4) at general inspection level II, single sampling, normal severity. That is the standard most glass and closure manufacturers already work to, and it gives you a defensible sample size that scales with lot size.

Here is the table I put in front of every supplier during qualification. These are the sample sizes at level II, normal inspection:

Lot size	Sample size (code letter)	Ac/Re at AQL 0.65	Ac/Re at AQL 1.0	Ac/Re at AQL 2.5
501 to 1,200	80 (J)	1 / 2	2 / 3	5 / 6
1,201 to 3,200	125 (K)	2 / 3	3 / 4	7 / 8
3,201 to 10,000	200 (L)	3 / 4	5 / 6	10 / 11
10,001 to 35,000	315 (M)	5 / 6	7 / 8	14 / 15
35,001 to 150,000	500 (N)	7 / 8	10 / 11	21 / 22

Ac/Re is accept number / reject number. If the cumulative defect count in the sample reaches the reject number, the lot is rejected. If it stays at or below the accept number, it passes. Read the table once and you will never accept “we pulled 10” again.

When to switch to tightened inspection

ISO 2859-1 has a switching rule most suppliers ignore. After two lots out of five consecutive lots are rejected on original inspection, you switch to tightened inspection (which drops the accept number by roughly one step and increases sample size for the same code letter). You stay on tightened until five consecutive lots are accepted, then return to normal. Write this into the spec. A supplier who never switches to tightened is not running a sampling system, they are running a pass/fail ritual.

Reduced inspection exists too, and I generally refuse it for oem vial product. Reduced inspection cuts sample size by 40% and raises the accept number. On a product where a single cracked vial can ruin a sequence, saving 80 pieces of inspection labor is not worth the risk.

What a real outgoing inspection record contains (and what a fake one contains)

This is where the difference between suppliers shows up fastest. A real outgoing QC record for an oem vial lot has seven things on it. If any are missing, the record is decoration.

Lot number and manufacturing date, traceable to the molding shift and the glass lot.
Quantity in the lot and the sample size drawn, with the ISO 2859-1 code letter stated explicitly.
Inspection level and AQL values used, by defect class, written on the record, not referenced to a document number nobody can find.
Defect counts by category (critical, major, minor) with the accept/reject decision for each.
Measured values, not just pass/fail, for the dimensional checks: body OD, height, neck ID, septum thickness, cap torque. At least 10 data points per dimension, with the mean and range.
Functional test results: leak test method (I use a 30-second pressure decay at 1.0 bar or a dye penetration per the supplier’s validated method), puncture count on the septum for multi-puncture closures (typically specified at 3 punctures for a 1.0 mm silicone/PTFE septum before reseal failure), and needle withdrawal force if the closure is a septa type.
Inspector ID, date, and signature, plus the reference to the control plan revision.

What a fake record contains: a single “PASS” stamp, a lot number, and a date. Or the opposite failure: a 40-page report with every value nominal and no measured data, generated from a template. I have seen both. The tell is whether the dimensional section shows a range. Real glass has a range. A record where every body OD reads exactly 12.00 mm is a record that was typed, not measured.

The measurement that catches most suppliers

Ask for cap torque as a measured value on screw-thread vials. Most suppliers report it as pass/fail against a spec of “hand tight.” That is not a number. I specify 0.4 to 0.9 N·m applied by a calibrated torque tester, with the sample of 20 closures, and I want the mean and standard deviation on the record. A supplier who cannot produce that number is not measuring torque, they are feeling it. The first time I asked a new supplier for torque data, the response was a photo of a technician’s hand. That lot went back.

The same applies to septum thickness on the closure. A 1.0 mm nominal silicone/PTFE septum that runs 0.85 mm at the low end will leak on a 9 mm screw cap at autosampler temperatures above 40°C. Ask for the measured distribution, not the nominal. If the supplier sources closures separately (many do), the closure inspection belongs in the same outgoing record as the vial. The vial closures you pair with the body are half the defect surface.

Two field cases where the sampling plan found the defect the COA missed

Case 1: the short-fill that passed visual

A contract lab running EPA 8270 on a 12 x 32 mm screw vial reported intermittent low recovery on the internal standard, roughly 8% RSD worse than their historical baseline on a 20-injection sequence. The vial was not the first suspect. It was the last. But when they weighed 50 vials from the suspect lot against 50 from the prior lot, the mean fill mass was 4.2 mg lower and the range was 11 mg wide against 3 mg historically. The supplier’s outgoing record showed a visual check for fill level, which was done by looking at the vial against a backlight. That catches gross short fills, not a 4% deficit.

The fix was to add a gravimetric fill check to the outgoing plan: 20 vials weighed to 0.1 mg, mean and range reported, spec ±3% of nominal fill mass. The supplier resisted at first because it added 15 minutes per lot. It also eliminated the customer’s repeat issue on the next three lots. The lesson: a defect class that can only be caught by instrument belongs in the plan with an instrument, not a visual check. The same logic applies to any chromatography vial where fill volume affects the injection.

Case 2: the crimp skirt that passed dimensional but failed seal

A pharmaceutical QC group using 11 mm crimp vials on a headspace sampler started seeing 2 to 3 leak failures per 100 vials on their own leak test, against a supplier outgoing record showing zero critical defects. The supplier’s plan measured the crimp skirt OD and the finish height, both in spec. What the plan did not measure was skirt concentricity relative to the neck axis. Off-axis crimping produced a seal that held at room temperature but failed at the 80°C headspace incubation.

The measurement that identified it was a simple runout check: rotate the crimped vial on a V-block and measure indicator runout at the skirt. The failing vials ran 0.15 to 0.22 mm; the passing vials ran under 0.08 mm. The fix was a concentricity spec of ≤ 0.10 mm TIR added to the outgoing plan, checked on 20 crimped samples per lot after the supplier’s own crimping step. The supplier’s crimp tooling had drifted on one of four heads. That is a process defect the sampling plan surfaced only because the defect class existed on the sheet.

Writing the sampling plan into the purchase order, not the quality manual

The single most effective thing I have done on this topic is move the AQL plan from a quality manual appendix into the purchase order as a line item. When the sampling plan is a PO requirement, it becomes a commercial term. A rejected lot is a nonconformance with a return authorization, not a conversation. When it lives only in the quality manual, it gets negotiated at every shipment.

Here is the decision checklist I use when qualifying a new oem vial supplier’s outgoing QC:

Does the inspection record state the ISO 2859-1 code letter and sample size, or just “sampled”? Reject if just “sampled.”
Are critical, major, and minor defects listed separately with distinct AQLs, or lumped into “visual”? Reject if lumped.
Is there at least one measured value per critical dimension with mean and range? Reject if pass/fail only.
Is cap torque reported as a number in N·m on a calibrated tester? Reject if “hand tight.”
Is there a functional leak or seal test with a stated method and duration? Reject if “visual seal check.”
Is the record traceable to the molding shift and glass lot? Reject if lot number only.
Does the supplier switch to tightened inspection after two rejects in five lots? Reject if no switching rule.

Two rejects on that list and I do not qualify the supplier, regardless of price. The cost of a bad lot is not the lot. It is the failed sequence, the re-run, and the investigation, which on a 96-well plate run is a full day of instrument time. The oem vial unit price is a small fraction of that.

The outgoing record is a process document, and it should look like one

I want to close on the thing that separates a real QC system from a certificate mill: the outgoing record should show variation, not perfection. A record from a good supplier shows a body OD mean of 12.01 mm with a range of 0.12 mm across the sample. It shows cap torque at 0.62 N·m mean with a 0.09 standard deviation. It shows one minor defect on a 200-piece sample and an accept decision at AQL 2.5. That is what a measured process looks like. A record showing every value at nominal and zero defects on every lot is either a very mature process or a document that was not produced by measurement. In my experience, on the oem vial side, it is usually the second.

If you want to verify the functional side of the plan independently, the leak and seal methods I described track the same approach used in third-party analytical testing labs for container closure integrity, and the chromatographic impact of a bad vial shows up in the same place the instrument vendors document it, in the injection-to-injection precision on their own consumables guidance (Agilent, Thermo Fisher). For method-level context on why a vial defect propagates into a result, the EPA SW-846 methods and Restek’s technical notes are the references I point new buyers to.

The decision rule is simple. Specify ISO 2859-1 level II, normal severity, with AQL 0.65 critical, 1.0 major, and 2.5 minor, written into the purchase order. Require the outgoing record to show measured values with ranges, a calibrated torque number, and a functional seal test. Reject any lot whose record shows pass/fail only, and switch that supplier to tightened inspection the first time two lots in five are rejected. If they cannot produce that record, they are not sampling, and the zero on their certificate is not a measurement.

References & further reading

Related resources from our lab consumables network:

External references and standards: