ISO 2859-1 is the framework, and most vial suppliers are using it wrong

A 40,000-vial lot of 2 mL clear crimp-top vials failed outgoing inspection at 0.65% dimensional defect rate against AQL 0.40, and the batch record showed nothing wrong. The in-process SPC charts were clean. The forming station had held its setpoints. What changed was the sampling plan: the previous shipment had been inspected at General Inspection Level I with a sample size of 32, and the switch to Level II pulled 125 vials and exposed a crimp-seat diameter drift that had been running for three shifts. That is the argument for a real hplc vial AQL sampling plan and outgoing QC program in one paragraph. The defect was always there. The sample was too small to see it.

I have signed off on outgoing inspection records for years, and the number of buyers who ask “what AQL do you use?” is far higher than the number who ask “what is your sample size, and how did you classify the defects?” The second question is the one that decides whether a shipment is actually conforming or just statistically lucky. This article is about the second question.

The reference standard is ISO 2859-1 (the international twin of ANSI/ASQ Z1.4). It defines attribute sampling plans by lot size, inspection level, and AQL. It is a mature, well-tested framework, and it is also routinely misapplied in this industry in three specific ways.

The first misapplication is using a single AQL for all defect classes. ISO 2859-1 does not require a single AQL, and a vial has at least three defect classes with wildly different consequences. Applying AQL 1.0 across the board means you accept a 1% rate of a defect that puts glass in a needle.

The second is using General Inspection Level I on lots above 10,000. Level I is cheap and it is what many suppliers default to when the buyer has not specified. For a 50,000-vial lot, Level I gives a code letter of H (sample size 50) and Level II gives L (sample size 200). The difference between those two samples is the difference between catching a 1% defect and missing it.

The third is treating the outgoing inspection as a pass/fail gate rather than a defect-classification exercise. If your record only says “PASS,” it is not a record. It is a rubber stamp. A usable record shows the sample size, the number of defects by class, the accept/reject numbers, and the disposition.

Defect classification: critical, major, minor, and the numbers I actually use

Before you can pick an AQL, you have to decide what a defect is. This is the part that takes the most negotiation with a supplier, because a lax classification is how a supplier hides a real problem inside a passing record. Here is the classification I use, and the AQLs that go with it.

Two things about this table. First, critical defects get a zero acceptance number regardless of sample size. That is not a statistical claim, it is a policy. A vial with a fragment in the cavity is not a vial with a probability of harm; it is a vial that will damage a needle and possibly a column. One of those is one too many. Second, Major A at AQL 0.40 is my default for crimp and screw-thread vials because the failure mode is visible at the autosampler, which means the customer sees it, which means it becomes a complaint. Major B at 0.65 covers the defects that fail a leak test or shift retention time but do not stop the run.

If you are buying septa separately, the classification shifts. A septum that is the wrong durometer or has a pinhole is a critical defect in my book, because it fails silently. The Hplc Septa specification should be sampled at a tighter AQL than the vial body. I use AQL 0.25 on septa for critical attributes and 0.65 for dimensional.

Sample size is a function of lot size, inspection level, and the number of defect classes you are sampling for

ISO 2859-1 gives you a code letter from the lot size and inspection level, then a sample size from the code letter. The table below is the slice that matters for vial lots, using General Inspection Level II, which is what I specify unless a customer demands otherwise.

Read the accept numbers carefully. At AQL 0.40 with n=125, the accept number is 1. That means you accept the lot if you find one defect in the sample. That is not a typo and it is not a loophole; it is the arithmetic of attribute sampling. The protection comes from the combination of sample size and accept number, not from the AQL number alone. This is why a supplier who quotes “AQL 0.4” without telling you the sample size and accept number has told you nothing useful.

I also specify that the sample is drawn by stratified random selection across the lot, not from the top of the carton. The top of the carton is where the supplier’s own picker touched the vials, and it is not random. A proper draw uses a random number table or a simple randomizer to pick cartons, then vials within cartons. If the supplier cannot describe how the sample was drawn, assume it was not.

What the outgoing inspection record should actually contain

I have audited a lot of supplier records. The ones I trust share a structure. The ones I don’t share a slogan. Here is what a realistic record contains, in the order I want to see it.

• Lot identification and traceability. Lot number, manufacturing date, forming station ID, glass lot, septum lot, cap lot. If the septum lot is missing, the record cannot support a complaint about a leak.
• Sampling plan statement. The standard (ISO 2859-1), the inspection level (II), the AQLs by defect class, the sample size, and the accept/reject numbers. Not “AQL 0.65.” The full plan.
• Sample draw method. How cartons and vials were selected. A random number seed or a description of the stratified draw.
• Test methods and instruments. The dimensional gauge (with calibration date), the leak test method (pressure decay, dye penetration, or vacuum), the torque tester, the puncture tester. Each with a method reference.
• Results by defect class. Number of defects found, by class, with the accept/reject decision for each class. Not a single overall pass.
• Out-of-spec disposition. If a class failed, what happened to the lot. Rework, sort, scrap, or concession. A concession without a customer signature is a supplier deciding for you.
• Signatures and date. Inspector and QC reviewer. Two names, not one.

The single most common gap I see is the missing test method for the leak test. “Leak tested” is not a method. Pressure decay at 0.5 bar for 30 seconds with a 0.5 mbar threshold is a method. Dye penetration with a 0.1% methylene blue solution under vacuum for 60 seconds is a method. If the record does not name the method and the threshold, the result is not reproducible and not auditable.

The second most common gap is the missing septum and cap lot numbers. A vial is an assembly. The Vial Closures are the parts that fail most often at the customer, and a record that cannot trace them is a record that cannot support a root cause investigation.

Torque, puncture, and dimensional thresholds: the numbers that go in the record

An outgoing inspection record is only as good as the thresholds it tests against. These are the ones I write into a specification, and they are the ones a supplier should be able to reproduce.

• Crimp-seat OD: 11.00 mm ±0.15 mm for a standard 11 mm crimp vial. Measured with a calibrated digital micrometer at the seal seat, not the body. I reject at ±0.20 mm without argument because the crimper cannot compensate.
• Neck ID: 6.00 mm ±0.10 mm for a 2 mL vial. This is the dimension that governs needle entry and septum compression.
• Cap torque (screw thread): 0.9 to 1.4 N·m for a 9 mm screw cap, measured with a torque tester on a sample of 10. Below 0.9 you get leaks; above 1.4 you get septa that extrude or crack.
• Septum puncture count: 3 punctures minimum with a 22-gauge needle at 10 psi, no coring. A cored septum is a critical defect.
• Leak test: pressure decay at 0.5 bar, 30 s hold, ≤0.5 mbar drop. Or dye penetration, 0.1% methylene blue, 60 s vacuum. Either is acceptable; neither is optional.
• Glass defect in seal zone: visual at 10x, no chips, no checks, no seeds larger than 0.2 mm within 2 mm of the seal seat.

Note the GPI reference on the cap: a standard 9 mm screw cap follows the GPI 9 mm finish, and a 11 mm crimp follows GPI 11 mm. If the supplier cannot state the GPI finish in the record, the closure fit is unverified. That matters more than most buyers realize, because a cap that fits a non-conforming finish will still pass a torque test and still leak.

Field case 1: the crimp-seat drift that passed Level I

The symptom was a customer complaint about autosampler needle bends on a specific lot of 2 mL crimp vials. The customer had run about 3,000 vials and reported 14 bent needles, roughly 0.47%. That is inside a Major B AQL of 0.65, so on the surface the lot was conforming.

The measurement that identified the cause was a crimp-seat OD check on 50 retained vials from the same lot, using a calibrated micrometer. The mean was 11.09 mm with a range of 11.02 to 11.19 mm. The upper spec is 11.15 mm. The process was centered high, and the tail was out. The in-process SPC chart had been plotting the mean, which was in spec, and had not been plotting the individual values against the upper limit. The outgoing record used Level I (n=50) with an accept number of 1 at AQL 0.65, and the sample happened to miss the tail.

The fix was two-part. First, the supplier moved the crimp-seat dimension to a 100% gauge check at the forming station, because a 0.04 mm centering error is not something an attribute sample catches reliably. Second, the outgoing plan moved to Level II with n=125 and added a variable-data check on 10 vials per lot with a Cpk target of 1.33. The next three lots ran at a mean of 11.04 mm and Cpk 1.51. The customer complaint stopped.

Field case 2: the septum lot that passed incoming and failed at the needle

The symptom was a retention-time shift of about 0.06 min on an early-eluting peak, appearing on one lot of vials and not the next. The customer’s first suspicion was the column, which is the usual first suspicion and almost always wrong. The Agilent consumables documentation was checked, the column was fine, and the shift followed the vial lot.

The measurement that identified it was a septum durometer check on 20 septa from the suspect lot, compared with 20 from a known-good lot. The suspect lot ran at Shore A 48 ±3, the good lot at Shore A 55 ±2. The softer septum compressed more under the crimp, which changed the headspace volume and shifted the early peak. The outgoing record showed the septum lot number, which is why the investigation was possible at all. It also showed a “septum hardness: pass” line with no value and no method, which is why the defect had not been caught at outgoing inspection.

The fix was to add a durometer check to the outgoing plan as a Major A attribute, sampled at AQL 0.40, with the specification written as a numeric range and a method reference (ASTM D2240, Shore A, 1 s reading). The supplier’s septum vendor was also put on notice. The lesson is that a record with a lot number and no threshold is a record that documents the existence of a test, not its result.

The outgoing inspection checklist I hand to a new supplier

When I qualify a new vial supplier, I send this checklist and I ask for a completed example against a real lot. A supplier who cannot fill it in is not ready to ship to me, regardless of price.

• Sampling standard and inspection level stated in full (e.g., ISO 2859-1, Level II, normal inspection, single sampling).
• AQLs by defect class, with the accept/reject numbers for each, not just the AQL number.
• Sample size and how the sample was drawn (random or stratified, with method).
• Defect classification list with definitions and examples for critical, Major A, Major B, and minor.
• Test methods named, with instrument IDs and calibration dates for micrometer, torque tester, leak tester, and puncture tester.
• Lot traceability for vial body, septum, and cap, each with its own lot number.
• GPI finish stated for the closure (GPI 9 mm or GPI 11 mm).
• Variable-data results for the key dimensions (mean, range, Cpk) on a 10-piece sample, not just attribute pass/fail.
• Disposition rule for a failed class, including who signs a concession.
• Retention sample policy: how many vials are held, for how long, and where.

Two items on that list are the ones suppliers push back on. The variable-data results are the first, because attribute sampling at n=125 tells you whether the lot is acceptable and says almost nothing about whether the process is centered. A Cpk on the crimp-seat OD is what catches the drift in field case 1. The second is the retention sample. I want 50 vials held for 24 months from the lot, because that is the only way to investigate a complaint six months after shipment. A supplier who does not retain samples cannot investigate anything.

If you are sourcing the full assembly, the same logic applies to the closures and septa as separate line items. The Chromatography Products supply chain has enough variability in septa durometer and cap dimensions that buying them as a matched assembly from one supplier, with one outgoing record, is usually worth the small premium over piecemeal sourcing. The Hplc Vial specification should be the controlling document for the assembly, not three separate purchase orders.

For buyers working under regulatory methods, the outgoing record also has to survive an audit. If your method references a waste extraction protocol, the container requirements in U.S. EPA SW-846 and the QC documentation conventions in Restek’s technical resources are worth a read before you write the specification, because they shape what the record must contain. The Thermo Fisher chromatography sample handling documentation is a useful cross-check on closure compatibility for the autosampler you actually run.

What to do when the outgoing record fails

A failed outgoing inspection is not a disaster. A failed outgoing inspection that gets shipped anyway is. The disposition rules I write into a purchase order are short.

Critical defect found: lot is rejected, no rework, no concession. The supplier sorts 100% or scraps. There is no AQL that accepts a critical defect at any sample size, and a supplier who proposes a concession on a critical defect has told you what their quality system is worth.

Major A or Major B class failed: the supplier may 100% sort the lot and re-submit it for a new outgoing inspection at the same sample size. The re-submitted lot gets a new record, and the original failure stays in the record. Do not let a supplier overwrite the failure. The history is the point.

Minor class failed: the supplier may rework (relabel, re-carton) and re-inspect at the same plan. A minor failure that requires rework of more than 10% of the lot is a signal that the process, not the lot, needs attention.

Any concession requires a signed deviation from the buyer, with the defect class, the quantity, and the acceptance rationale. If the deviation is verbal, it does not exist.

The takeaway is a decision rule, not a summary. Specify the sampling plan by standard, level, AQL by defect class, and accept/reject numbers before you place the order. Require the outgoing record to name the method, the instrument, the lot numbers for body, septum, and cap, and the variable-data Cpk on the key dimension. Reject any record that says “pass” without a number behind it. The sample size is the cheapest insurance in the whole supply chain, and the suppliers who resist it are the ones who need it most.

References & further reading

Related resources from our lab consumables network:

• Hplc Vial at hplcvialfactory.com
• Chromatography Products at chromatographyforsale.com
• Hplc Septa at hplcsepta.com
• Vial Closures at vialsclosures.com

External references and standards:

• Agilent chromatography consumables and columns
• Thermo Fisher chromatography and sample handling
• U.S. EPA SW-846 test methods for waste and materials
• Restek technical resources for chromatography