Last March a lot of 200,000 2 mL clear vials, 12 × 32 mm, 9 mm screw thread, failed outgoing inspection at the crimp-and-cap station because 14 of 315 sampled units had septa seated more than 0.4 mm proud of the cap skirt. The customer’s fill line caught it three days later as a needle-height fault on a 7693A tower. The AQL was 0.65. Fourteen defects in a 315-piece sample is roughly 4.4% defective, more than six times the acceptance number for that plan. Nobody had written the sampling plan down. The inspector was pulling “about a tray” per pallet and eyeballing caps under a desk lamp. That is the gap this article closes: what an autosampler vial AQL sampling plan actually looks like on paper, how the defect classes are split, and what a real outgoing QC record contains when a customer audits it.

Most vial manufacturers run a three-class defect scheme borrowed from ISO 2859-1 (the ANSI/ASQ Z1.4 equivalent in the US). The classes are not arbitrary. They map to what happens downstream.

• Critical (Class A): defects that make the vial unusable or unsafe in the autosampler. Wrong thread pitch, cracked neck, septum missing, cap not crimped, glass shard in the cavity, dimensional excursion on the 32 mm height beyond ±0.5 mm. AQL 0.65, and I would argue for 0.25 on crack and shard. Zero acceptance is tempting but statistically unworkable at realistic lot sizes unless you sample 100%.
• Major (Class B): defects that cause a functional failure the customer will notice but that the instrument may tolerate once. Septum proud of skirt by more than 0.3 mm, cap torque below the minimum, visible glass seam at the shoulder, dimensional excursion on OD between 11.6 and 12.0 mm outside tolerance but inside 0.3 mm. AQL 1.0.
• Minor (Class C): cosmetic. Faint scuff on the body, print smudge on the label, minor bubble in the glass that does not affect the seal. AQL 2.5 or 4.0 depending on the customer’s specification.

The choice of AQL is a business decision, not a statistics decision. AQL 0.65 on critical means you accept that up to 0.65% of the lot may be defective before the plan rejects. If your customer runs 96-well plates at 1,000 injections a week, 0.65% is 6.5 bad vials per week. That is tolerable if the failure mode is a cosmetic scuff and intolerable if the failure mode is a needle strike. This is why critical and major cannot share an AQL.

Sample size follows lot size, not gut feel

Under ISO 2859-1 general inspection level II, a lot of 200,000 pieces maps to sample size code letter P, which is 800 pieces. That is a lot of vials to pull, measure, and cap-check per lot. Most vial plants run at general inspection level I or S-3 for routine outgoing, and reserve level II for first article and customer PPAP submissions. At level I, code letter N, sample size 500. At S-3, code letter J, sample size 80. The acceptance numbers shift accordingly.

My working rule: level II, sample size 315 (code letter M, lot 35,001 to 150,000) for the first three lots of any new SKU or new mold, then level I at 200 pieces per lot for steady-state production, with a mandatory switch back to level II if any lot in the last 10 was rejected. That tightening rule is what makes a sampling plan a plan rather than a ritual.

Defect classification has to be written against the instrument, not the drawing

The drawing says the vial is 32.0 ± 0.5 mm tall. The instrument says the needle penetrates to a fixed depth and the gripper closes to a fixed span. A vial at 32.4 mm may pass the drawing and still sit 0.4 mm high in the tray, which changes the septum puncture depth by the same amount. If your major defect list stops at the drawing tolerance, you are shipping units that fail in the field without ever appearing on your QC sheet.

I classify against three references in this order:

1. GPI 9 mm thread finish (the Glass Packaging Institute standard that the 9-425 screw thread follows). Thread major diameter, pitch, and the neck ring position are all defined here. A cap that threads on but bottoms out 0.2 mm short of the seal face is a major defect even though the thread gauges pass.
2. Instrument interface dimensions from the autosampler manufacturer. Agilent publishes tray and gripper dimensions for the 7693A and 1260 series; Thermo publishes equivalent data for the TriPlus and AS/TS lines. These are the numbers that decide whether the vial sits in the tray.
3. Customer drawing, which is often the least complete of the three.

The defect class assignment then follows: any excursion that touches reference 1 or 2 is critical or major. Any excursion that only touches reference 3 is major or minor depending on whether the customer’s own incoming inspection will catch it.

What gets measured, and with what

An outgoing inspection record that passes audit has specific instruments named. In my plants the station carries:

• Digital caliper, 0.01 mm resolution, for OD, height, and neck OD. Gauge R&R must be under 10% of tolerance.
• Thread plug gauge to GPI 9 mm, go/no-go, for thread form.
• Torque tester with a 0-100 N·cm range, for cap application torque. Target 18 to 22 N·cm for a 9 mm screw cap on a PTFE/silicone septum. Below 15 N·cm the cap backs off in transport; above 28 N·cm the septum deforms and coring risk rises.
• Seated-height gauge, a simple go/no-go block, for septum proud-of-skirt. The limit is 0.3 mm above the skirt top.
• Visual standard board with photographed reference defects at 10× magnification, re-signed by QC lead every six months.
• Leak tester (pressure decay) on a 30-piece subsample per lot, for closure integrity at 1.0 bar for 30 seconds.

That last one catches the failure mode that visual inspection cannot: a cap that looks perfect but has a septum compressed on one side and not the other. A pressure decay test at 0.5 bar differential will flag it in under a second per vial.

The outgoing inspection record is a document, not a checkbox

When a pharma customer audits your plant, they will ask for the outgoing inspection record for a specific lot number and expect it within 30 minutes. If you cannot produce it, the audit fails regardless of how good the vials were. The record I keep contains, at minimum:

• Lot and sub-lot identifiers, with mold cavity traceability. Every vial from a 32-cavity mold carries a cavity number on the base. The record lists which cavities were in the sample and which, if any, produced the defects found.
• Sampling plan reference: standard (ISO 2859-1), inspection level (e.g. II), sample size code letter, sample size, AQL by class, acceptance and rejection numbers.
• Measurement data, not pass/fail. The 315 individual height readings, not just “conforms.” A customer’s statistician will compute Cpk from them and compare against your certificate of analysis. If your CoA says Cpk 1.67 and the raw data says 1.42, you have a credibility problem.
• Defect log, with defect class, cavity number, and a photograph of each defect type found. Even if the lot is accepted, the log shows what was found and dispositioned.
• Instrument identification: caliper serial number, calibration due date, torque tester serial, leak tester serial. Each must be traceable to a calibration record.
• Inspector name and date, with the inspector’s training record referenced by ID.
• Disposition: accept, accept with deviation, reject, rework. If rework, the rework record is attached.

This is roughly 4 to 6 pages per lot. It is not optional. A lot that fails has a longer record, not a shorter one, because the rejection, the root cause, and the corrective action all attach to it.

Field case: the cavity that drifted

In 2022 a 32-cavity mold on a 2 mL clear vial line started producing a slow drift in neck OD on cavity 17. Over 11 lots the neck OD moved from 8.42 mm to 8.58 mm, still inside the ±0.15 mm drawing tolerance, so every lot passed. The outgoing record carried the raw data but nobody was plotting it. At lot 12 the cap torque on cavity-17 vials dropped below 15 N·cm because the thread major diameter was now at the high edge of the GPI band and the cap was riding the crest rather than the flank. A customer’s fill line flagged it as a cap-loosening complaint on a 40-vial tray. The fix was a mold insert replacement on cavity 17 and a new rule: the outgoing record includes a control chart of neck OD by cavity, evaluated against a 6-lot rolling mean, with a mandatory action trigger at 60% of tolerance consumed. That is a sampling plan that does something.

Sampling frequency is a function of process stability, not a fixed schedule

If your outgoing inspection is “sample 200 pieces at the end of the shift,” you are sampling the wrong thing. The sample has to be pulled across the production window, not from the last hour’s output. I use a stratified pull: 8 time points across the run, 25 pieces per point at level I, and 16 time points, 20 pieces per point at level II. That catches a startup drift or a mid-shift mold temperature excursion that a single end-of-run sample would miss entirely.

Frequency also depends on the process capability you have demonstrated. A line running at Cpk 1.67 on all critical dimensions can move to a reduced inspection scheme under ISO 2859-1 switching rules: after 10 consecutive lots accepted at normal inspection, switch to reduced. Reduced inspection uses roughly 40% of the sample size. Any single rejection switches back to normal, and two rejections in five lots switches to tightened, which uses roughly 1.6× the sample size. These switching rules are the whole point of the standard. If you are not using them, you are running a fixed sample plan and calling it AQL.

What “reduced” does not mean

Reduced inspection does not mean skipping the leak test or the septum-seat check. Those are functional tests with their own sampling rationale, and they stay at full frequency regardless of the AQL switching state. What changes is the dimensional sample size. I have seen plants reduce the leak test to one vial per lot under “reduced inspection” and then ship a lot with a 3% leak rate. The switching rules apply to the attribute sampling plan, not to the functional tests. Keep those separate.

What the customer’s incoming inspection will do with your record

The customer will do three things with your outgoing record. First, they will verify the sampling plan matches the one in the supply agreement. If your agreement says AQL 0.65 critical and your record shows AQL 1.5, the lot is technically nonconforming even if it passed. Second, they will compare your raw data against their own incoming inspection data on the same lot. A systematic offset of more than 0.05 mm on any dimension means one of you has a gauge problem. Third, they will compute their own Cpk from your raw data and compare against the value on your CoA. Any discrepancy triggers a supplier corrective action request.

The practical consequence: your outgoing inspection record is not a shipping document. It is the input to the customer’s supplier qualification. Treat it as such. If a dimension is trending, show the trend. If a defect was found and dispositioned, show the disposition. Records that show only pass/fail invite the customer to run their own inspection at level II, and if their sample is larger than yours, they will find defects you missed.

For reference on the analytical side of what the customer does with the vial once it passes, the Agilent chromatography consumables and columns and Thermo Fisher chromatography and sample handling resources document the instrument-side tolerances that the vial has to meet. The Restek technical resources library has useful application notes on how vial geometry affects injection reproducibility, which is the end-use context your QC plan is protecting.

A worked decision checklist for a new lot

Here is the checklist I hand to a new QC inspector on day one. It is not a template for the record; it is the decision sequence that produces the record.

Two things about this checklist that are not obvious. Step 3 measures all 320 because dimensional data is cheap and the control chart needs the full set. Step 7 is a functional test that overrides the AQL switching state: if it fails, the lot is held regardless of what the attribute plan says. A lot can pass AQL and still fail the leak test, and the leak test wins.

Field case: the septum that passed AQL and failed the customer

A lot of 100,000 9 mm screw caps with PTFE/silicone septa passed outgoing inspection at AQL 0.65 critical, 1.0 major. The customer rejected it on incoming inspection for coring on a 6890 autosampler at 200 injections per vial. The outgoing record showed septum thickness at 1.02 mm, inside the 1.00 ± 0.10 mm spec. What the record did not show was septum durometer. The incoming silicone had drifted from 45 Shore A to 52 Shore A between lots, and the harder septum cored at 180 punctures instead of the 400 the customer expected. The fix was a durometer check added to the outgoing plan at 5 pieces per lot, with a 40-50 Shore A acceptance band, and a supplier requirement that durometer be reported on every incoming silicone certificate. That is a defect class that did not exist on the original plan because nobody had written it down. The autosampler vial specification is only as good as the defect classes on the inspection sheet.

Where the plan meets the rest of the supply chain

The outgoing inspection plan does not exist in isolation. It has to align with the incoming inspection your customer runs and with the incoming inspection you run on your own components. A 9 mm cap that passes your outgoing torque test at 18-22 N·cm may still fail the customer’s incoming torque check if their gauge is calibrated differently or if they measure after a 48-hour settling period. The only way to close that gap is to agree on the measurement method in the supply agreement, including the settling time, the gauge type, and the torque range.

The same applies to the vial dimensions. If your outgoing plan measures height with a caliper and the customer measures with an optical comparator, you will disagree on borderline units. Write the method into the plan. For headspace applications, the headspace vial plan adds a crimp-dimension check at 20 mm and a seal-integrity check at 1.5 bar, because the crimp geometry is what fails under headspace pressure. For GC applications, the GC vial plan adds a residual-volume check because the bottom geometry drives the injection volume repeatability. Those are different plans for different products, and they should not share a defect list.

The regulatory frame is also worth stating. If the vial is used in an SW-846 method, the container is part of the analytical batch, and the U.S. EPA SW-846 test methods for waste and materials documentation defines what that means for the customer. The outgoing inspection record is the evidence that the container met the specification the customer’s method assumed. That is the whole point of the plan.

The decision rule

Write the plan before the first lot ships, not after the first complaint. Set critical at AQL 0.65 with a sample size of 315 at inspection level II for new SKUs and 200 at level I for steady state. Classify defects against GPI 9 mm and the instrument interface, not against the customer drawing alone. Measure all sampled units on dimensions, subsample for torque and leak, and keep the raw data. Switch inspection levels using the ISO 2859-1 rules, but never reduce the functional leak test. If a lot fails, the record gets longer, not shorter. That is the whole discipline: a sampling plan that produces a document a customer can audit, and a defect list that reflects what actually fails in the autosampler.

References & further reading

Related resources from our lab consumables network:

• Autosampler Vial at autosamplerhplc.com
• Vial at vials-filters.com
• Headspace Vial at headspacevialsupplier.com
• Gc Vial at gcvialfactory.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