A 12,500-piece lot of 20 mL headspace vials came back from the receiving dock with a 0.9% capping defect rate, and nobody could tell me which sublot it came from because the outgoing inspection record listed one line: “visual: OK.” That is the failure mode this article is about. An voa vial AQL sampling plan is not a signature on a packing slip. It is a statistical statement about a lot, and if the record cannot reconstruct which samples were pulled, which defect class each fell into, and what the accept/reject arithmetic was, the plan did not happen.
I have written and audited these plans on both sides of the transaction, as an OEM supplier shipping to contract labs and as a buyer receiving from three glass plants. What follows is the version I would defend in a customer audit, with the numbers I actually use.
ISO 2859-1 is the base document, but you have to pick the inspection level deliberately
The sampling tables everyone cites come from ISO 2859-1 (the ANSI/ASQ Z1.4 equivalent in the US). The standard gives you a matrix: lot size on one axis, inspection level and AQL on the other, and a sample size code letter that resolves to n and to Ac/Re (accept number / reject number). The part people get wrong is the inspection level. General inspection level II is the default and it is too loose for a consumable that fails by the thousand at the autosampler.
For voa vials I run general inspection level II for cosmetic and dimensional attributes and level S-4 or level III for the closure-critical attributes (seal integrity, cap torque, septum puncture). That split is a decision, not a compromise. Cosmetic defects are cheap to find statistically; a single leaking vial in a 100-position tray can cost a full sequence and a re-prep. You want tighter sampling where the consequence is a failed run.
The lot-size bands that matter in practice:
- 1,201 to 3,200 pieces: code letter K at level II gives n = 125. At level III, n = 200.
- 3,201 to 10,000: code letter L at level II gives n = 200. At level III, n = 315.
- 10,001 to 35,000: code letter M at level II gives n = 315. At level III, n = 500.
- 35,001 to 150,000: code letter N at level II gives n = 500. At level III, n = 800.
Note that the sample size is capped by the table, not by lot size. A 100,000-piece lot at level II is inspected on 500 pieces. That is the whole point of acceptance sampling: you are buying a defined producer’s and consumer’s risk, not certainty. If your customer expects certainty, sell them 100% automated vision inspection and price it accordingly. Do not pretend an AQL plan is 100% inspection.
Three defect classes, and the AQL I assign to each
The AQL number is a quality level, expressed as percent nonconforming, that the plan is designed to accept most of the time. Lower AQL means tighter. The mistake is applying one AQL to every attribute. Classify first, then assign.
Critical defects: AQL 0, Ac = 0
A critical defect is one that creates a safety, regulatory, or total-loss condition. For a voa vial that means: a pinhole or channel in the septum that leaks under the headspace pressurization cycle, a cap that is not crimped at all, glass with a through-crack, or a vial that fails the extractable-screen because the wrong liner was installed. Critical gets AQL 0 with Ac = 0. Any single critical defect rejects the lot, full stop. Do not average it against the sample size. One leaking septum in 315 is a rejected lot and a supplier corrective action request.
Major defects: AQL 0.65
Major defects make the vial unusable or force rework but do not create a safety event. Examples: cap not seated to the crimp dimension, septum not centered (offset greater than 0.5 mm), dimensional out-of-tolerance on the critical sealing diameter, missing or illegible lot code, torque outside the 8 to 12 in-oz band for a crimp cap. I use AQL 0.65 for majors. At n = 200 that gives Ac = 3, Re = 4. At n = 315, Ac = 4, Re = 5. Those numbers are from the ISO 2859-1 single sampling table and you should verify them against the edition your QA system references before you print them on a form.
Minor defects: AQL 1.5 to 2.5
Minor defects are cosmetic or nuisance-level: a scuff on the glass body that does not affect the seal, a label print that is faint but legible, a mold seam line. I set AQL 1.5 for a premium line and 2.5 for a commodity line. This is where most of the argument with the glass plant happens, because the plant wants to lump everything into “minor” and pass the lot.
The table below is the plan I would put in a specification document. Adjust the AQLs to your customer’s contract, but do not adjust the classification logic.
| Defect class | Examples for a voa vial | AQL | n at lot 3,201-10,000 (level II) | Ac / Re |
|---|---|---|---|---|
| Critical | Leaking septum, uncrimped cap, through-crack, wrong liner material | 0 | 200 | 0 / 1 |
| Major | Cap torque out of 8-12 in-oz, septum offset > 0.5 mm, sealing OD out of tolerance, missing lot code | 0.65 | 200 | 3 / 4 |
| Minor | Body scuff, faint print, mold seam, cosmetic chip outside seal zone | 1.5 (premium) / 2.5 (commodity) | 200 | 7 / 8 or 10 / 11 |
Verify those Ac/Re pairs against your own copy of the standard. I have seen two suppliers cite different accept numbers for the same AQL and sample size because one was reading level II and the other level III. The form should state the inspection level explicitly so the arithmetic is reproducible by an auditor.
What actually goes on the outgoing inspection record
This is the part that separates a real plan from a rubber stamp. A defensible outgoing QC record for a voa vial lot has, at minimum, the following fields. If any one is missing, the record cannot survive a customer audit and it cannot support a recall decision.
- Lot and sublot identifiers, tied to the glass plant’s forming shift and the capping station ID. One lot number covering three shifts is not traceable.
- Date and time of sampling, plus the sampler’s name or ID.
- Sample size n, inspection level, and AQL per defect class, stated as numbers, not “per procedure.”
- Sampling method: stratified by tray position or by capping head, not a grab from the top of the carton. If you sample from one carton you are testing one carton.
- Defect counts by class, with the accept/reject decision shown as arithmetic (e.g., majors: 2 found, Ac = 3, accept).
- The measurement data behind the pass: cap torque readings (in-oz), septum offset, critical dimensions with the gauge ID and calibration due date.
- Seal integrity test result: pressure-decay or dye penetration, with the test pressure and dwell time.
- Retained samples: quantity, storage location, retention period. I keep 30 pieces per lot for 24 months.
- Disposition: released, reworked, or rejected, with the signature and date of the person with release authority.
The record should also carry the certificate of analysis reference and the material lot numbers for glass and septum. If a customer later reports a ghost peak, the first question is which septum lot went into the vial, and the answer has to be on the record, not reconstructed from memory. This is the same traceability logic that governs the voa vial supply chain generally, applied at the shipment gate.
Seal integrity is not a visual attribute, so do not inspect it visually
I have watched suppliers “inspect seal integrity” by squeezing the cap and looking at it. That is not a test. A crimp that looks perfect can leak at 15 psi, and a crimp that looks slightly off can hold. The seal is a mechanical interference between the septum, the glass finish, and the aluminum crimp, and it has to be measured.
Two methods I accept on an outgoing record:
Pressure decay. Pressurize the headspace to a defined test pressure (I use 15 psi for a 20 mm crimp finish), isolate, and measure the pressure drop over a 10-second dwell. A pass is a drop below a threshold you establish from a known-good baseline. Run it on a sample of 30 from the lot, separate from the AQL sample, because it is a destructive or semi-destructive test.
Dye penetration. Immerse the capped vial in a dye bath under vacuum, then inspect for dye ingress at the seal. Slower, but it catches the channel defects that pressure decay can miss at low differential.
Neither test belongs in the AQL attribute sample. They are separate verification tests with their own sample size and their own accept criteria, and the record should show them as such. The septa supplier’s own incoming data on liner thickness tolerance feeds directly into whether these tests pass, so the two records should cross-reference.
Case 1: the lot that passed AQL and still leaked in the field
Symptom. A contract lab running EPA Method 8260 reported that roughly 1 in 60 vials from a recent lot showed a low response on the internal standard, consistent with a slow leak during the 14-day hold. The outgoing record showed a clean pass: 200 sampled, zero majors, zero criticals.
What the measurement found. We pulled the retained samples and ran pressure decay. Mean pressure drop was within spec, but the distribution had a right tail: about 2% of vials dropped more than 3x the median. The AQL sample had caught none of them because the attribute inspection was visual plus a single torque reading per vial, and torque was in band on every one. The root cause was a capping head that had drifted on crimp height by about 0.15 mm across one shift. Torque was fine; crimp height was not. The AQL plan could not see it because crimp height was not a measured attribute on the form.
Fix. Added crimp height as a measured major attribute with a tolerance of 0.1 mm, sampled 10 per capping head per shift, and moved seal integrity to a separate pressure-decay test on 30 pieces per sublot. The next lot shipped with a right tail of zero. The lesson is not that AQL failed. The lesson is that an attribute has to be on the form before the plan can catch it.
Case 2: the sublot that the record could not isolate
Symptom. A buyer rejected a 30,000-piece shipment because 4 of 500 sampled vials had septum offset beyond 0.5 mm. Majors: 4 found against Ac = 4 at n = 500, level II. Technically an accept at the boundary, but the buyer’s spec was tighter and they wanted the affected pieces segregated.
What the measurement found. The outgoing record listed one lot number for the whole 30,000. The capping operation had run across two stations. We could not tell which station produced the offset septa, so we could not segregate. We ended up 100% sorting the entire lot at our cost, which is exactly the outcome a proper sampling plan is supposed to prevent.
Fix. Split lot numbering by capping station and shift, so a lot never spans more than one station. The outgoing record now shows the station ID and the sample is stratified across tray positions. The next rejection could be isolated to a sublot of about 6,000 pieces and sorted in a day.
This is where the vial septa and closure data has to be linked to the lot record, because septum offset is a function of both the septum and the cap, and blaming one without the other is guesswork.
Torque is a measured attribute, not a feel
For a crimp cap, the parameter is crimp height and crimp diameter, not rotational torque, but for a screw-thread voa vial the relevant number is application torque. I specify 8 to 12 in-oz for a 20 mm screw cap with a PTFE-faced silicone septum, measured with a calibrated torque tester on a sample of 10 per capping head per shift. Below 8 in-oz you get leakage under headspace pressure; above 12 in-oz you cold-flow the septum and lose the seal on the second puncture.
Puncture count is the other number that belongs on the record, because it is a design limit, not a QC finding. A typical silicone septum tolerates 3 to 5 punctures before the reseal degrades; a PTFE-faced septum is good for a similar count if the needle is 23-gauge or smaller. If your customer’s method punctures twice (once to add, once to sample), the septum has margin. If the method punctures five times, the septum is at its limit and the AQL plan should be sampling for coring and reseal, not just for offset. The HPLC vial closure specs and the GC headspace specs differ here, so state which one the lot is built to.
A decision checklist for the shipment gate
Before a voa vial lot leaves the building, the following should be true. This is the checklist I hand to a new QC lead on day one.
- Lot number is unique to one capping station and one shift, and the record shows both.
- Inspection level and AQL are stated per defect class on the record, not referenced to a procedure.
- Sample size matches the ISO 2859-1 table for the lot size and the stated level. Recompute it; do not trust the form.
- Sample is stratified across tray positions, not pulled from one carton.
- Cap torque or crimp height measured on 10 per capping head, with gauge ID and calibration date.
- Seal integrity tested separately on 30 pieces, with test pressure and dwell recorded.
- Critical defects: zero accepted, any one rejects the lot.
- Retained samples: 30 pieces, labeled, stored, retention period stated.
- Certificate of analysis cross-references glass lot and septum lot.
- Disposition signed by someone with release authority, not the person who ran the sample.
If any line is blank, the lot does not ship. That rule is easier to enforce than it sounds, because the checklist is short enough that a QC lead can run it in under an hour per lot.
The decision rule
Set the plan from the consequence of failure, not from the lot size. Critical defects get AQL 0 and Ac = 0 with a separate seal-integrity test on 30 pieces; majors get AQL 0.65 at inspection level II or III depending on the customer’s tolerance for rework; minors get AQL 1.5 to 2.5. Then write the record so it can reconstruct which sublot, which capping head, which gauge, and which accept/reject arithmetic produced the release. If the record cannot answer those four questions, the plan is not an AQL plan, it is a signature. For the underlying material and dimensional specifications, the Agilent GC consumables reference, the Thermo Fisher chromatography sample-handling documentation, the EPA SW-846 methods, and the Restek technical library are the references I cite when a customer asks why a given attribute is on the form.
References & further reading
Related resources from our lab consumables network:
External references and standards: