Pick the AQL from the cost of the failure, not from habit

A 20,000-piece lot of 2 mL clear crimp vials left our dock with a signed inspection sheet showing zero defects. Three weeks later a customer returned 1,400 units because the septa sat 0.4 mm proud of the crimp skirt on roughly 7% of them, enough to break the autosampler’s gripper alignment on every tenth tray. The outgoing record said “visual: OK.” That phrase cost us a credit note and a re-inspection of the whole lot. The problem was not the vials. It was the sampling plan behind the signature. If your outgoing QC sheet cannot tell you which AQL level you applied, which defect class a proud septum belongs to, and how many pieces you actually pulled, you are not inspecting, you are guessing with a clipboard.

This is a working account of how I build a vial AQL sampling plan for outgoing QC, the numbers I put on the sheet, and the defect classification that makes the plan mean something. It assumes you already make or buy the product and need the shipment gate to hold.

Most vial plants default to AQL 1.0 for major defects because that is what the old sheet said. That number is a policy choice disguised as a technical one. The right way to set it is to ask what a defective unit does downstream.

I split vials into three failure consequences and assign AQL accordingly:

• Function-stopping defects (a vial that cannot be crimped, seated, or pierced): AQL 0.65 or tighter. A single bad vial in a 100-position autosampler tray aborts the run.
• Method-affecting defects (dimensional drift that changes headspace volume, a septum with the wrong durometer): AQL 1.0. The vial still works, but the chromatogram shifts.
• Cosmetic defects (a faint scratch on the body, a print smudge): AQL 2.5 or 4.0. Nobody’s data changes.

These map to the ISO 2859-1 general inspection levels. I use General Inspection Level II for routine outgoing lots and drop to Level I only when the lot is small and the process has been stable for at least 20 consecutive lots with no major findings. I never use Level III on a routine basis; it triples the sample size and the marginal detection gain rarely justifies the labor. The exception is a first article or a post-changeover lot, where I will run Level III once and then step back to II.

The AQL you write on the sheet is a promise about the process average, not a guarantee about the lot. At AQL 1.0 and a sample of 200, you accept the lot if you find 5 or fewer major defects. That is a producer’s risk of about 5% and a consumer’s risk that a lot running at 4% defective will still pass. If a customer’s incoming inspection is tighter than your outgoing plan, you will ship lots they reject. Align the two numbers before the first shipment, not after the first complaint.

Sample size follows the lot, and the table is the only honest source

I have seen QC sheets where the inspector pulled “about 30” from a 50,000-piece lot. At General Level II, a lot of 35,001 to 150,000 calls for a code letter N sample of 500 pieces. Thirty pieces cannot detect a 1% defect rate with any confidence. The sample size is not a judgment call; it comes off the ISO 2859-1 table for the lot size and inspection level you selected.

Here is the working extract I keep taped inside the QC cabinet. It covers the lot ranges we actually run:

Read the accept/reject columns carefully. At AQL 0.65 with a code letter G sample of 32, the acceptance number is zero. One defect and the lot fails. That is a hard gate, and it is why AQL 0.65 belongs on function-stopping defects only. Applying it to a cosmetic class will fail half your lots for scratches nobody cares about.

Two operational rules I enforce on the floor. First, the sample must be drawn across the production stream, not off the top of the carton. I pull from at least five positions in the lot and from at least three different trays or boxes. Second, the sample is drawn after final packaging is complete, because a vial that passes bare can still fail once the septum and cap are on. The vial is only half the assembly; the closure is the other half, and outgoing inspection has to see the finished unit.

Classify defects before you count them

A sampling plan without a defect classification is a coin flip with a signature. The inspector needs a written list of what counts as critical, major, and minor, with a measurable threshold for each. Here is the classification I use for a 2 mL crimp vial with a PTFE-lined butyl septum.

Critical defects (AQL 0.065, zero acceptance)

• Any crack, chip, or star check visible under 10x magnification on the neck or body.
• Septum punctured, torn, or missing.
• Wrong septum material in the lot (a silicone septum in a butyl order).
• Foreign particulate inside the vial cavity visible at 10x.

Critical defects get a zero acceptance number regardless of sample size. A single cracked vial in a tray will shed glass into the autosampler needle path, and that is a run-killing event.

Major defects (AQL 1.0)

• Neck OD outside ±0.15 mm of nominal (for a 11 mm crimp finish, nominal 11.00 mm).
• Crimp skirt height outside ±0.20 mm.
• Septum seated more than 0.20 mm proud or recessed relative to the skirt.
• Body height outside ±0.25 mm.
• Visible seam or mold line crossing the sealing surface.

Minor defects (AQL 2.5)

• Scratches on the body that do not reach the glass interior.
• Print misregistration on the label or graduation mark.
• Color variation in the cap or septum within an agreed tolerance band.

The dimensional thresholds are not arbitrary. A 0.20 mm proud septum is roughly the point at which a standard autosampler gripper starts to misalign. A 0.15 mm neck OD excursion is where a crimp tool begins to leave an uneven seal. If you cannot measure to those tolerances, the classification is fiction, and you need a go/no-go gauge rather than a caliper and a hope. The headspace vial line runs the same logic with tighter neck tolerances because the headspace volume is part of the method.

What a realistic outgoing inspection record actually contains

The signed sheet is the only artifact that survives a customer audit. I have seen sheets with a single “PASS” stamp and a date. That is not a record. A defensible outgoing inspection record for a vial lot has, at minimum, these fields:

• Lot or batch number and the manufacturing date range it covers.
• Quantity in the lot and the packaging configuration inspected.
• Sampling standard and level (ISO 2859-1, General Level II, single sampling, normal inspection).
• Sample size and the code letter it came from.
• AQL applied per defect class, with the accept and reject numbers written out.
• Defect counts by class, not a total. A lot with 0 critical, 2 major, and 9 minor is a different animal from one with 2 critical and 0 major.
• Measurement data for the dimensional majors: at least the min, max, and mean for neck OD, body height, and septum seat, with the gauge or instrument ID.
• Inspector name and date, and the name of the person who released the lot.
• Disposition: accept, reject, rework, or sort, with the rework record referenced.

I also record the inspection level switch history. If the plan moved from normal to tightened after two consecutive rejected lots, the sheet says so. That switch is the whole point of ISO 2859-1 switching rules, and auditors look for it. A plant that has never switched to tightened inspection in three years is either perfect or not running the plan.

Field case: the proud septum that passed visual

A crimp vial line was shipping lots with a 6 to 8% rate of septa seated 0.3 to 0.5 mm proud. The inspection sheet showed zero major defects because the inspector was checking “septum present” with a thumbnail, not measuring seat height. The measurement that caught it was a simple go/no-go ring gauge sitting on the crimp skirt: if the septum touched the gauge, it failed. The fix was twofold. We added septum seat height to the major class with a 0.20 mm limit, and we moved the check from the end of the line to immediately after the crimping station, where the operator could see the trend. Outgoing reject rate on that defect dropped to under 0.5% within two lots.

Measure the dimensions that the method actually depends on

Not every dimension on the drawing belongs on the outgoing sheet. Outgoing QC is a gate, not a full dimensional layout. I measure the features that change the analytical result or the mechanical fit, and I leave the rest to first article inspection.

For a 2 mL crimp vial the outgoing measurement set is:

• Neck OD: 11.00 mm nominal, tolerance ±0.15 mm, measured with a calibrated snap gauge at 90° to the seam.
• Crimp skirt height: measured with a depth gauge, ±0.20 mm.
• Body height: 32.00 mm nominal for a standard 2 mL, ±0.25 mm.
• Septum seat height: go/no-go ring gauge, 0.20 mm limit.
• Pierce force: on a sample of 10, using a 22-gauge needle at 300 mm/min, target 4 to 8 N. This is the check that catches a durometer drift in the septum before the customer’s autosampler does.

The pierce force check is the one most plants skip and the one that catches the most expensive escapes. A septum that measures correctly but has drifted 5 Shore A harder will pass every dimensional check and then blunt needles and shed cores in the field. The GC vial line runs a pierce force check on every outgoing lot for exactly this reason.

If you buy rather than mold, ask the supplier for the pierce force data on the septum lot, not the vial lot. The septum is a separate incoming material and its durometer is where the variability lives. For the reference on how the closure interacts with the container, the U.S. EPA SW-846 test methods set container performance expectations that a vial supplier should be able to trace to.

Field case: the lot that failed on neck OD, and the switch that caught it

A 60,000-piece lot of 2 mL vials came off a mold that had been running for 11 days without a tooling check. Outgoing inspection at Level II pulled 500 pieces and found 9 major defects against an accept number of 7 at AQL 1.0. The lot was rejected. The dimensional data showed the neck OD had drifted from a mean of 11.01 mm to 11.14 mm, with a max of 11.19 mm, right at the upper tolerance limit. The root cause was wear on the neck ring insert, which had gone 0.08 mm oversize.

The interesting part was the switching rule. The previous lot had already been rejected for a different major defect (septum seat). Under ISO 2859-1, two consecutive rejected lots move the plan to tightened inspection, which uses a smaller acceptance number for the same sample size. The third lot, drawn under tightened inspection, failed on 4 major defects against an accept number of 3. That third failure is what triggered the tooling teardown. Without the switching rule, the third lot would have passed under normal inspection and shipped. The fix was a tooling wear limit of 0.05 mm on the neck ring, checked every 5,000 pieces, plus a permanent move to tightened inspection for 5 lots after any tooling change.

The lesson generalizes. The sampling plan is not just a sample size and an AQL. The switching rules are what turn a static gate into a feedback loop, and they are the part most plants omit because the sheet template never had a field for them.

The outgoing plan is not a substitute for process control

I want to be blunt about one thing. Outgoing AQL sampling catches escapes; it does not fix a process. If your outgoing reject rate is running at 2 to 3% on majors, the sampling plan is telling you the process is out of control, and the answer is upstream, at the forming station and the crimping station, not a bigger sample. The plan’s job is to make sure a bad lot does not ship while you fix the process.

That said, the outgoing plan is the last line, and it has to be real. A plan that pulls 32 pieces from a 100,000-piece lot because “that is what we always did” is theater. A plan that classifies a proud septum as cosmetic is theater. A plan with no accept/reject numbers written on the sheet is theater.

When you specify a vial for a regulated method, match the outgoing plan to the method’s tolerance for dimensional drift. A method that is sensitive to headspace volume needs the body height and neck OD checks at the tight end. A method that runs on a standard autosampler needs the septum seat and pierce force checks. The Agilent chromatography consumables and columns documentation and the Thermo Fisher chromatography and sample handling resources both publish autosampler dimensional envelopes that your outgoing tolerances should sit inside. The Restek technical resources are a good cross-check for septum and liner compatibility if you are qualifying a new closure.

For a broader look at how the container and closure interact across the product range, the reagent bottle line runs the same outgoing logic at a larger scale, and the defect classification carries over almost unchanged.

The decision rule I would put on the wall

Set the AQL by failure consequence (0.65 for function-stopping, 1.0 for method-affecting, 2.5 for cosmetic), take the sample size from ISO 2859-1 General Level II for the lot size, write the accept and reject numbers on the sheet before the inspector touches the lot, classify every defect against a written threshold with a number on it, and record the switching history. If a lot fails, the sheet should tell you which class failed and by how much, not just that it failed. If your outgoing record cannot answer “which AQL, which sample size, which defect class, which measurement,” the shipment gate is not a gate. Fix the sheet before you ship the next lot.

References & further reading

Related resources from our lab consumables network:

• Vial at vials-filters.com
• Headspace Vial at headspacevialsupplier.com
• Gc Vial at gcvialfactory.com
• Reagent Bottle at reagentbottleforsale.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