Pick the AQL from the defect class, not from the customer’s purchase order

The outgoing inspection record for lot 24-117 showed 0.8% AQL on the critical defect class and a clean pass, yet the customer rejected the shipment on arrival: 14 of 500 vials would not seat in a 2 mL autosampler tray because the crimp skirt measured 0.18 mm out of round. The inspector had gauged the neck OD, not the skirt. That is the whole problem with cod vial AQL sampling plans in one lot. You can pass a sample and still ship a defect, because the plan only protects you to the extent the defect class, the gauge, and the sample size match what the customer actually does with the part. Outgoing QC is not a rubber stamp on the end of the line. It is a statistical contract, and most factories write it badly.

This piece is about what that contract should say, how to pick the AQL and the sample size, how to classify defects so the numbers mean something, and what a realistic outgoing inspection record contains. It assumes you already make or buy cod vial product and know the dimensional stack. The question here is narrower: given a finished lot, how many do you pull, what do you measure, and what do you write down.

Most of the sampling plans I audit have one AQL for the whole product. That is wrong on its face. A cod vial has at least three populations of characteristics, and they fail at different rates with different consequences. Collapsing them into one number either over-inspects the cosmetic stuff or under-inspects the functional stuff.

I classify defects into three tiers and assign a separate AQL to each:

• Critical (AQL 0.065 to 0.10): anything that makes the vial unusable in the method or a safety issue. Wrong septum face material, missing PTFE liner, neck ID outside the crimp spec so the seal leaks, glass that will not survive autoclave. These are typically zero-acceptance in the sample regardless of AQL, because a single leak in a 96-position tray ruins a run.
• Major (AQL 0.65 to 1.0): functional but recoverable. Skirt out of round enough to bind in a tray, thread or crimp geometry that needs rework, dimensional drift on the 11 mm or 13 mm crimp, cap torque below spec. The customer notices these and calls you.
• Minor (AQL 1.5 to 2.5): appearance and handling. Scuffs on the glass body, label registration, minor cosmetic on the cap. The customer may not even log these.

The AQL you write down should be traceable to a standard. ISO 2859-1 is the general acceptance sampling standard most buyers reference, and it is the one I default to. For critical defects I use a tightened plan and often a c=0 acceptance number even at small sample sizes, because the math of accepting one critical defect in a sample of 50 is ugly. For major and minor I use the normal inspection tables at the AQL above, switching to tightened when the process shows a trend on the control chart. If a customer hands you a purchase order with a single “AQL 1.0” line and no defect classification, push back. That line is not a specification. It is a placeholder.

Sample size is a function of lot size and inspection level, and it is smaller than you think

Under ISO 2859-1 general inspection level II, a lot of 501 to 1,200 pieces calls for a sample of 80. A lot of 1,201 to 3,200 calls for 125. A lot of 10,000 still only calls for 200. People are surprised by this. They assume a big lot needs a big sample. It does not, because the standard is built around the acceptable quality level, not around proportional coverage.

Here is the working table I keep on the wall for a cod vial line running general level II, single sampling, normal inspection:

Two things to read off that table. First, at AQL 0.65 on a 500-piece lot you accept on one defect and reject on two. That is a tight plan, which is what you want for a major functional defect. Second, the acceptance number at AQL 2.5 for a 200-piece sample is 10. That is a lot of cosmetic defects before you reject, and it is why minor defects should almost never trigger a lot hold on their own. If your QC lead is holding lots over label registration at a 2.5 AQL, the plan is doing what it was told. The AQL is the problem.

For critical characteristics I do not use the table at all. I pull a fixed sample (typically 32 to 50 depending on lot size) and require zero defects. The reason is practical: the consequence of a single wrong septum face in a 96-position tray is a failed run for the customer, and no AQL above zero is defensible for that.

Defect classification has to be written down before the inspector pulls the first vial

I have watched two inspectors grade the same vial differently because the classification sheet said “skirt roundness: major” without a number. One inspector passed a skirt that bound in the tray. The other failed it. Both were following the sheet. The sheet was the defect.

A usable defect classification for a cod vial names the characteristic, the gauge, the measurement, and the threshold. It looks like this:

• Neck OD (critical): measured with a calibrated pin gauge or optical comparator at the sealing surface. Accept if within the drawing tolerance for the crimp spec (11 mm or 13 mm). Reject any out-of-tolerance part.
• Crimp skirt roundness (major): measured as runout on a dial indicator with the vial on a mandrel, or by seating in a go/no-go tray gauge. Threshold set at the autosampler tray clearance minus 0.05 mm. The 0.18 mm out-of-round in my opening example would have failed this if it had been measured.
• Septum face material (critical): verified by lot certificate and by a spot check of the liner color and durometer. A silicone-only septum where PTFE-faced was specified fails here, not at the customer.
• Cap torque (major): measured with a calibrated torque tester on a sample of capped units. For a screw cap on a 9 mm or 10-425 neck I set the target band and tolerance from the cap manufacturer’s spec, typically 0.6 to 1.0 N·m, and reject anything outside. For crimp caps, torque does not apply; crimp diameter and seal integrity do.
• Glass body defects (minor): scuffs, seeds, minor checks that do not affect the seal or the autosampler fit. Graded against a visual standard, not against a mood.

If the classification sheet has an adjective where a number should be, the plan is not ready. Fix the sheet before you fix the inspector.

The gauge is the plan: what a cod vial outgoing inspection actually measures

You cannot sample what you cannot measure. The most common failure I see in outgoing QC is not a bad AQL. It is a gauge that does not match the characteristic. The opening example is the canonical case: neck OD was gauged, skirt roundness was not, and the lot shipped with a major defect that the plan never had a chance to catch.

The gauges a cod vial line needs, at minimum:

• Pin gauges or an optical comparator for neck ID and OD. Calibrated, with a calibration sticker and a due date on the bench.
• A go/no-go tray gauge machined to the autosampler tray pocket dimensions plus clearance. This is the only gauge that answers the actual customer question: does it seat? I have seen lines with a $4,000 comparator and no tray gauge. The tray gauge is cheaper and more relevant.
• A torque tester for screw caps, calibrated against a known weight or a certified torque standard.
• A leak or seal test for critical lots: pressure decay or dye penetration on a sample of capped vials. For crimp-sealed vials this is the only way to catch a bad crimp that looks fine.
• A durometer or a liner verification method for septum face material, backed by the supplier’s certificate of analysis.

For reference on what the downstream instruments expect, the Agilent GC columns and accessories documentation and the Thermo Fisher chromatography and sample handling pages both publish autosampler dimensional expectations. Match your tray gauge to those, not to a drawing from 2019.

What a realistic outgoing inspection record contains

I have seen inspection records that are one line: “Lot 24-117, sampled 80, passed.” That record is worthless in a dispute. A record that holds up in an audit or a customer complaint has, at minimum, these fields:

• Lot or batch number and the manufacturing date range it covers.
• Sampling plan reference: the standard (ISO 2859-1), the inspection level (II), the AQL per defect class, and the sample size actually pulled.
• Defect classification sheet version used, with a revision date. If the sheet changed, the record says which version applied.
• Gauge IDs and calibration due dates for every gauge used. A measurement from an out-of-calibration gauge is not a measurement.
• Raw measurements, not just pass/fail. The actual neck OD readings, the actual torque values, the actual roundness runout. Pass/fail hides drift. Raw numbers let you see a lot that is trending toward the limit before it crosses.
• Defect counts by class and the accept/reject decision per class, not a single overall decision.
• Inspector ID and date, and a second signature for any lot dispositioned as use-as-is or reworked.
• Disposition of nonconforming units (scrapped, reworked, returned) with quantities.

If you are buying rather than making, ask for this record before you place the first order. A supplier who cannot produce a record with raw measurements and gauge IDs is not running an outgoing inspection. They are running a hope. The EPA SW-846 test methods that drive much of the cod vial demand do not specify a sampling plan for the vial itself, but they do specify data quality, and a lot with no traceable inspection record is a data quality problem waiting to surface in a lab audit.

Two field cases where the plan passed and the shipment failed

Case 1: the skirt that was never gauged

Symptom: a customer running a 96-position autosampler reported that roughly 3% of vials would not seat in the tray and had to be manually reseated, slowing the run and occasionally jamming the tray. The outgoing record showed a clean pass at AQL 0.65 on the major class.

Measurement: we pulled 200 vials from the retained sample and ran them through a tray gauge machined to the autosampler pocket dimensions. 7 of 200 bound, a 3.5% defect rate, which is far above the 0.65 AQL acceptance. The inspection record had gauged neck OD only. The skirt roundness was never measured, so the major class had no data behind it.

Fix: added the tray gauge to the major class, set the threshold at tray clearance minus 0.05 mm, and re-ran the retained sample. The 3.5% defect rate was there all along. The plan had been passing it because the gauge did not match the characteristic. We also tightened the AQL on that characteristic to 0.65 with a c=0 acceptance for the first three lots after the change, then relaxed to the table once the process showed it was stable.

Case 2: the torque band that was too wide

Symptom: a customer running headspace GC reported intermittent leaks on a screw-cap cod vial, showing up as low recovery on a subset of vials. The outgoing record showed torque within spec on every sampled unit.

Measurement: we pulled the customer’s retained vials and measured torque on 50 of them with a calibrated tester. The values ranged from 0.4 to 1.3 N·m against a spec of 0.6 to 1.0 N·m. The outgoing record had been measuring torque on the capping machine setpoint, not on the finished vial. The machine was drifting within its own tolerance, and the finished vials were outside the spec band. The record said “within spec” because the record was measuring the wrong thing.

Fix: moved torque measurement to the finished vial, sampled 10 per hour during the run, and set the control chart limits at 0.7 and 0.9 N·m inside the 0.6 to 1.0 spec. The tighter control band caught the drift before the finished vials left the spec. The customer’s leak rate dropped to zero on the next three lots. The lesson is not about torque. It is that an inspection record that measures the process instead of the product is not an outgoing inspection.

A decision checklist for the plan before the first lot ships

Before any lot leaves the building, the plan should answer yes to all of these. If it answers no to any, the lot is not ready to ship under that plan.

• Does every defect class have its own AQL, and is the AQL traceable to ISO 2859-1 or an equivalent standard?
• Is the sample size derived from lot size and inspection level, not from a habit?
• Does every characteristic on the classification sheet have a number, a gauge, and a threshold, with no adjectives standing in for measurements?
• Is every gauge calibrated, with an ID and a due date on the record?
• Does the record capture raw measurements, not just pass/fail?
• Is there a tray gauge, and is it matched to the autosampler the customer actually uses?
• Is there a seal or leak test on the critical class for crimp-sealed lots?
• Is there a second signature for any use-as-is or rework disposition?
• Is there a tightened inspection trigger tied to the control chart, and is it written down?

If you buy rather than make, the same checklist applies to your supplier’s record. A supplier who cannot show you the tray gauge and the raw torque values is a supplier you are sampling blind. The consumables side of this is well documented by vendors like Restek, whose technical resources cover the closure and septum behavior that drives most of these defects, and it is worth reading their notes before you write your classification sheet.

Recommendation: a working plan you can adopt this quarter

Here is the plan I would put on a cod vial line today, and the parameters I would defend in an audit:

• Standard: ISO 2859-1, single sampling, general inspection level II, normal inspection, switching to tightened on a control chart trend.
• Critical class: AQL 0.065, c=0 acceptance, fixed sample of 32 to 50 depending on lot size. Characteristics: neck ID/OD, septum face material, seal integrity on crimp lots.
• Major class: AQL 0.65. Characteristics: skirt roundness via tray gauge, cap torque on finished vials (0.6 to 1.0 N·m for a typical screw cap, control limits 0.7 to 0.9), crimp diameter.
• Minor class: AQL 2.5. Characteristics: glass body cosmetics, label registration.
• Record: the nine fields above, with raw measurements and gauge IDs.
• Retained sample: keep the pulled sample for the shelf life of the lot, labeled with the lot number and inspection date, so a customer complaint can be traced back to the actual units inspected.

If you are sourcing rather than making, the same plan is the specification you hand your supplier, and the retained sample is what you ask them to hold. The vial insert and 2 mL vial lines that feed the same autosampler have the same trap: a plan that gauges the wrong characteristic passes the lot and fails the customer. The broader HPLC consumables category has this problem across the board, and it is why I write the classification sheet before I write the AQL.

The decision rule is this: an outgoing inspection plan is only as good as the gauge behind each characteristic and the raw number on the record. If a characteristic has no gauge and no number, it is not in the plan, no matter what the AQL line says. Fix the gauge and the record first, then argue about whether the AQL should be 0.65 or 1.0. The argument over the AQL is the easy part. The gauge is the hard part, and it is the part that decides whether the shipment passes at the customer or only at your dock.

References & further reading

Related resources from our lab consumables network:

• Cod Vial at codvial.com
• Vial Insert at hplcinsert.com
• 2Ml Vial at 2mlsamplevial.com
• Hplc Consumables at hplcconsumable.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