AQL 1.0 at level II is the default, and the default is usually wrong for

A 20,000-piece lot of 250 µL glass inserts sat in a bonded warehouse for eleven days because the outgoing inspection record showed AQL 1.0 at general inspection level II, and the inspector had pulled 200 pieces from a single master carton. The buyer’s quality engineer caught it in the paperwork review: the sample was drawn from one of forty inner boxes. Statistically the inspection said almost nothing about the lot. That is the failure mode this article is about. Not the insert, not the glass, but the sampling plan and the outgoing QC record that is supposed to prove the lot is shippable. If you are buying inserts in volume, the AQL sampling plan and outgoing QC documentation is the only thing standing between you and a warehouse full of parts you cannot use.

AQL 1.0 at level II is the default, and the default is usually wrong for inserts

Most insert suppliers will quote ISO 2859-1, general inspection level II, AQL 1.0 for major defects and AQL 2.5 for minor defects. That is the textbook answer, and it is what you will find in a supplier’s quality manual. It is also the wrong plan for a part where a 0.15 mm dimensional drift turns into a sampler pickup error.

Here is the arithmetic. Under ISO 2859-1, a lot of 20,001 to 150,000 pieces at level II requires a sample size code letter of M, which means 315 pieces. At AQL 1.0, the acceptance number is 7 and the rejection number is 8. So the plan accepts a lot where up to 7 of 315 pieces fail. That is a defect rate of roughly 2.2% at the acceptance boundary, and the actual outgoing quality limit (the AOQL) for that plan is around 1.8%. If your insert is going into an autosampler that runs 96 positions a day, a 1.8% defect rate means you are pulling a bad insert out of the tray roughly every other day.

For vial inserts specifically, I would not accept AQL 1.0 for the critical defect class. I would not accept AQL 2.5 for the major class either. Here is how I classify defects and set the levels, and I will defend each number.

Defect classification for a glass insert

Three classes, and the classification is driven by what the defect does downstream, not by how it looks.

• Critical (AQL 0.065 or zero acceptance): anything that can physically stop the sampler or damage the needle. That means a chipped or cracked top rim that can shed glass into the vial, a bore obstruction (a fused or partially closed taper), and a base that is not flat to within 0.20 mm across the seating face. These are the defects that produce a hard fault, not a soft one. I use a zero acceptance number (c=0) on critical defects regardless of the AQL table, which means the plan is really “sample and reject on first finding.”
• Major (AQL 0.65): dimensional non-conformance outside the drawing tolerance on the outside diameter, the overall length, or the taper angle. A 250 µL insert that should be 5.0 ± 0.1 mm OD and 31.0 ± 0.3 mm overall, measured at 0.65, will still seat in most 2 mL vials but will start to bind in a tight one. A taper angle off by more than 0.5° is major because it changes the residual volume and the needle guidance.
• Minor (AQL 1.5 to 2.5): cosmetic issues that do not affect function. Slight fire-polish variation, a faint scuff on the outer wall, a bubble in the glass that is fully enclosed and under 0.3 mm. These get AQL 2.5 and nobody loses sleep.

Note that I put dimensional non-conformance in major, not critical. That is a deliberate choice and it is the one place where I will argue with a customer’s incoming inspection spec, because the customer’s drawing often calls the OD a critical characteristic when the sampler does not. The sampler cares about the seating face and the bore, not the OD to the third decimal. This is the same distinction that makes the vial insert a mechanical part first and a glass part second.

Sampling from the wrong place is worse than sampling too little

The eleven-day hold in the opening scene was not a sample-size problem. It was a stratification problem. The inspector pulled 200 pieces from one master carton out of forty. The defect that triggered the hold, a batch of inserts with a 0.4 mm short overall length, was confined to two inner boxes from a single forming cavity. A sample drawn from one carton would have missed it with high probability, and in this case it did.

My rule for inserts: sample across cartons, not within one. If a lot is 40 cartons, draw at least one piece from each carton before you draw a second piece from any carton. The ISO 2859-1 sample size tells you how many pieces, not where to get them. The “where” is a separate instruction and it belongs in the inspection record as a written procedure, not as an inspector’s judgment call.

What a realistic outgoing inspection record contains

This is the part most suppliers get wrong, and it is the part you should demand. A one-page certificate of analysis with a checkmark next to “visual: pass” is not an outgoing inspection record. A real record has these fields, and I have never seen a supplier produce all of them without being asked:

• Lot identification: the forming date, the cavity number or cavity group, the glass batch (if the supplier tracks it), and the carton range sampled. If the supplier cannot tell you which cavity made the sample, the sample is not traceable.
• Plan statement: the standard (e.g., ISO 2859-1), the inspection level, the sample size code letter, the sample size, and the AQL per defect class. Written out, not implied.
• Sampling instruction: “one piece per carton, cartons selected by random number table from the 40-carton lot,” or equivalent. This is the line that catches the warehouse hold problem.
• Measurement results, not pass/fail: the actual OD, length, and taper readings for each measured piece, or at minimum the min, max, mean, and RSD%. A pass/fail column hides drift. A mean that has crept from 31.02 to 31.18 mm over three lots is a warning even if every piece is in tolerance.
• Gauge identification and calibration date: the micrometer or optical comparator used, with its calibration due date. A dimensional result from an uncalibrated gauge is not a result.
• Defect tally by class: critical, major, minor, with the acceptance and rejection numbers stated next to the tally.
• Disposition: accept, reject, or accept under deviation with the deviation number and the approving signature.

If your supplier’s record is one page with three checkmarks, ask for the raw measurement sheet. If they cannot produce it, the AQL plan is theater. The same standard applies whether you are buying inserts or the 2 mL vial they sit in, and the record format should be identical so the two can be filed together.

Attribute sampling is not enough for a dimensional part

Attribute sampling (go/no-go) is cheap and fast, and for cosmetic defects it is fine. For dimensional defects on a glass insert, it is a poor fit because the part’s function depends on a continuous variable, not a binary one. A go/no-go gauge tells you the OD is between 4.9 and 5.1 mm. It does not tell you the lot mean has shifted to 5.08 mm, which is still passing but now has almost no margin against the upper limit.

My recommendation: run a variables sampling plan on the three dimensions that matter (OD, overall length, taper angle) and an attribute plan on everything else. Variables sampling under ISO 3951-1 gives you the same protection with a smaller sample. For a lot of 20,000 at inspection level II and AQL 0.65, the variables plan needs a sample of about 125 pieces instead of 315, because it uses the measured mean and standard deviation rather than a defect count. That is a real saving in inspection time, and it produces the mean and RSD% data that the record needs anyway.

The catch is that variables sampling assumes the characteristic is normally distributed. Glass forming is not perfectly normal, and a short overall length from a worn mold tends to produce a left tail. So I use variables sampling for the OD and length, and I keep an attribute check on the taper angle because taper failures are bimodal (either the angle is right or the tooling is worn, with little in between). This is the kind of decision that separates a real QC plan from a copied one, and it is why I would rather see a supplier’s sampling rationale than their certificate.

Two field cases where the sampling plan, not the part, was the problem

Case 1: the RSD that passed every lot and still failed in the field

Symptom: a pharma customer running a 96-well autosampler reported intermittent pickup errors, roughly one in 400 injections, with no pattern by operator or method. The inserts were in tolerance on every certificate.

Measurement that identified it: we pulled the retained samples from three consecutive lots and measured the seating face flatness on an optical flat. The lot means were fine, but the RSD% on flatness had gone from 4.2% to 11.8% across the three lots, and the max reading had moved from 0.12 mm to 0.19 mm against a 0.20 mm limit. The attribute plan (flat/not flat against the 0.20 mm limit) had accepted every lot because no single piece exceeded the limit. The distribution had widened, and the tail was now close enough to the limit that a small number of pieces were marginal in the sampler.

Fix: we changed the outgoing plan for that characteristic from attribute to variables, added an RSD% acceptance limit of 8% on seating face flatness, and put a control chart on the lot means. The pickup errors stopped within two lots. The part had not changed. The sampling method had been blind to the drift.

Case 2: the AQL plan that accepted a lot with a fused bore

Symptom: a contract lab received a 50,000-piece lot of 150 µL inserts and found, on first use, that about 1 in 200 had a partially fused bore that would not pass a 0.5 mm needle. The supplier’s certificate showed AQL 1.0, accept.

Measurement that identified it: we asked for the raw tally. The inspector had used a visual check under 2x magnification and had classified the fused bores as minor cosmetic defects (a “faint internal mark”). The defect class was wrong, so the AQL was wrong. At AQL 2.5, the plan accepted up to 14 minor defects in a 315-piece sample. The fused bores were a critical defect that should have triggered rejection on the first finding.

Fix: we rewrote the defect classification to put bore obstruction in critical with c=0, added a go/no-go pin gauge (0.5 mm) as a 100% check on the first article and a sampled check thereafter, and required the supplier to photograph any internal mark above 0.2 mm. The reworked lot passed. The lesson is that the AQL number is meaningless without the defect classification, and the classification is where most suppliers are sloppy.

The decision checklist I hand to a new supplier

This is the sheet I actually use. It fits on one page and it forces the supplier to commit to numbers before the first purchase order.

Two lines on that table do most of the work. The c=0 on critical defects is non-negotiable for me, because the cost of a single fused bore reaching a sampler is a stopped run, not a rejected insert. And the RSD% reporting is what turns a certificate into a trend you can act on.

Torque, puncture count, and the checks that belong in the outgoing record

Dimensional and visual checks are the core of the plan, but two functional checks belong in the outgoing record for any insert that ships with a cap or septum, and they are the ones suppliers skip because they take time.

Torque: if the insert ships in a capped 2 mL vial, the cap torque matters because an over-torqued cap deforms the septum and an under-torqued cap leaks. The outgoing check should be a sampled torque measurement, typically 1.2 to 1.5 N·m for a 9 mm screw cap on a 2 mL vial, with the gauge and the sample size in the record. I have seen a lot rejected because the torque mean was 1.7 N·m, well above the target, and the septa were visibly deformed. That is a major defect and it should have been caught before shipment.

Puncture count: for a septum, the outgoing record should state the puncture count the lot is rated for, typically 3 to 5 punctures for a standard silicone/PTFE septum before reseal integrity degrades. If your supplier does not state a puncture count, the septum is unspecified, and an unspecified septum is a defect waiting to happen. This is the kind of thing that belongs in the HPLC consumables spec sheet, not in a verbal assurance.

Both checks are cheap. A torque gauge is a few hundred dollars and a puncture test is a few minutes per sample. The reason they get skipped is that the outgoing plan was written for the insert alone, not for the insert as it ships. If the insert ships inside a capped vial, the outgoing plan has to cover the assembly.

What to do when the outgoing record and the field failure disagree

Sooner or later a lot passes outgoing inspection and fails in the field. The record is your only tool, and how you use it determines whether you find the cause in a week or a quarter. My sequence:

• Pull the retained samples first. If the 20-piece retention set from the lot shows the defect, the outgoing plan missed it and the sampling method is the problem. If the retention set is clean, the defect happened after inspection (handling, shipping, storage) or the plan sampled the wrong stratum.
• Check the sampling location. Compare the cartons in the field failure against the cartons in the record. If the failed cartons were never sampled, you have a stratification problem, not a part problem.
• Re-measure the characteristic with a calibrated gauge. Do not trust the field measurement until you reproduce it. A 0.1 mm difference between two micrometers is common and it changes the conclusion.
• Compare the lot mean against the previous three lots. Drift shows up in the means before it shows up in the defect count. This is the single most useful thing in the record and the reason I insist on mean and RSD% rather than pass/fail.

If you need a second opinion on the gauge or the method, the technical resources at Restek and the sample handling documentation from Thermo Fisher are worth a look, and the instrument makers publish their own sampler tolerances that you can check the insert against. For anything going into environmental work, the container and handling requirements in U.S. EPA SW-846 set a floor on what the outgoing record has to demonstrate. And if you are sourcing the whole assembly, a chromatography supplier who can show you the raw inspection sheet is worth more than one who shows you a certificate.

The decision rule

Set the outgoing plan from the defect’s downstream consequence, not from a default AQL table. Critical defects (bore, rim, seating face) get c=0 and a pin gauge. Major dimensional defects get AQL 0.65 with variables sampling and mean/RSD% reported per lot. Cosmetic defects get AQL 2.5 and nothing more. Sample across cartons before you sample within one, keep 20 retained pieces per lot for 24 months, and refuse any certificate that shows a checkmark where a measurement belongs. If a supplier cannot produce the raw tally and the sampling location, the AQL plan is not a plan, it is a form, and the lot it “passed” is a lot you have not actually inspected.

References & further reading

Related resources from our lab consumables network:

• Vial Insert at hplcinsert.com
• 2Ml Vial at 2mlsamplevial.com
• Hplc Consumables at hplcconsumable.com
• Chromatography Supplier at chromatographysupplier.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