Last spring a lot of 18,000 crimped 2 mL vials sat on a dock in New Jersey waiting for a release decision. The customer’s incoming inspector had already pulled 50 pieces and found two with septum push-through, which put the lot at roughly 4% defective on a visual attribute. Under a 2ml vial AQL sampling plan at General Inspection Level II, AQL 1.0, the accept number for n=125 is 3. Two defects in 50 tells you almost nothing. The question was whether the outgoing inspection record from the supplier could carry the weight, and it could not, because the record said “visual: pass” and nothing else. That is the gap this article is about: how to build an outgoing QC sampling plan for 2 mL vials that actually resolves a dispute at the dock.

I have written and audited these plans on both sides of the transaction, as a molder shipping vials and as a buyer receiving them. The plans that work share a structure. The plans that fail are usually not wrong about AQL. They are wrong about defect classification, about sample size relative to the defect rate that matters, and about what gets written down.

Pick the inspection level from the defect rate you can tolerate, not from habit

Most suppliers default to General Inspection Level II because that is what the ANSI/ASQ Z1.4 table puts in front of you. Level II is a reasonable default for cosmetic attributes. It is the wrong level for the attributes that stop a run.

Here is the arithmetic that decides it. At Level II, lot size 10,001 to 35,000, the sample size is n=315 under normal inspection. At AQL 1.0 the accept number is 5 and the reject number is 6. That plan will accept a lot with a true defect rate of about 2.8% roughly 10% of the time. If a 2.8% coring rate would cost you a rerun on a 96-well autosampler plate, the plan is too loose. If the attribute is a scuff on the label area that nobody sees, it is too tight and you are burning inspector hours.

So the first decision is not the AQL. It is which attributes get which inspection level. My working split for 2 mL vials:

• Critical defects (septum breach, cap not seated, wrong liner material, glass crack through wall): 100% inspection on the critical feature, or a tightened plan at Level III with AQL 0.065 and c=0. Do not run a critical attribute through an AQL table at 1.0 and call it controlled.
• Major defects (dimensional out of tolerance on neck OD, crimp height, septum thickness; visible particulate inside the cavity): Level II, AQL 0.65.
• Minor defects (cosmetic scratches, faint lot-code print, mold seam flash under 0.1 mm): Level I, AQL 2.5, or skip-lot after 10 consecutive accepted lots.

The c=0 point matters and gets argued about. A zero-acceptance plan at n=125 has an actual consumer risk curve that is steeper than the AQL table suggests, and some suppliers will push back saying it is statistically unfair. They are right that it is harsh at low defect rates. They are wrong that it is inappropriate, because a critical defect on a 2 mL vial is not a quality attribute you are negotiating, it is a containment failure waiting to happen. If the septum is breached, the sample is gone and the run is gone.

Defect classification is where most plans quietly fall apart

I have read maybe 200 outgoing inspection records from vial suppliers. The single most common weakness is not the sample size. It is a defect list that classifies everything as “appearance” and then passes the lot on a visual scan under office lighting.

A workable classification for a 2 mL vial has to be tied to function, not to how the part looks. Here is the version I use, with the measurement that decides each one:

The neck OD gauge is worth a note. A 2 mL vial neck is nominally 11.0 mm at the crimp ring, and the tolerance matters because the crimper closes on it. If the neck runs 0.15 mm over, the crimp will not seat fully and you get a leak that only shows up after the vial is in the autosampler. I have seen that failure present as intermittent peak-area loss of 8 to 12% on the back half of a sequence, and the cause was a shipment of vials with a neck OD at the top of the tolerance band. A go/no-go ring gauge at the dock catches it in ten seconds per vial. A caliper reading catches it in thirty and depends on where the operator puts the jaws.

The liner material check is the one that saves the most money. A 2 mL vial with a PTFE/silicone septum and one with a butyl septum look identical in a bin. The butyl one will bleed siloxanes and give you ghost peaks at m/z 73, 147, 207, 281 in GC-MS. If the outgoing record does not include an FTIR or DSC spot check against a retained reference, you are trusting the molder’s resin intake, which is a different (and already-published) topic. At the dock, an FTIR check on 8 pieces per lot is cheap insurance.

How to size the sample when the defect you fear is rare

The AQL table is built for consumer risk, not for the rare catastrophic defect. If your concern is a 0.2% coring rate, no practical AQL plan will find it. At n=315 and c=0 you have about a 47% chance of catching a 0.2% defect rate. At n=125 it is 22%.

This is where the sample size decision has to split into two tracks. The AQL plan handles the population of common attributes. The rare critical attributes get a separate, targeted plan built around the physics of the failure. For coring, that means a puncture rig: 10 punctures per vial with a 22G needle at 90° to the septum, then a leak check. Run it on 32 vials per lot. That is 320 punctures per lot, and it will surface a coring problem at a rate the AQL table never will, because you are amplifying the exposure per piece.

The same logic applies to seal integrity. A weight-loss test (I have written elsewhere about why it is the honest test) run on 10 vials per lot over 72 hours at 40°C will catch a crimp problem that a visual inspection misses entirely. Fold that into the outgoing record as a lot-level attribute, not a per-piece one.

What a realistic outgoing inspection record contains

This is the part buyers ask about most and get the least. A record that says “AQL 1.0, Level II, PASS” is not a record. It is a claim. Here is what I would want to see, and what I put in front of customers when I am on the shipping side.

• Lot identity and traceability. Lot number, molding date, resin lot, septum lot, cap lot, and the machine/cavity set. Cavity-level traceability matters because a single bad cavity can produce a stripe of defects that a random sample dilutes.
• Plan citation. The standard (ANSI/ASQ Z1.4 or ISO 2859-1), the inspection level, the AQL per defect class, and the sample size actually drawn. Not the planned sample size, the drawn one.
• Defect tally by class. Counts of critical, major, and minor defects found, with the accept/reject numbers for each. If a lot was rejected on first sample and resampled, the record shows both draws. Resampling without disclosure is the most common integrity failure I see.
• Measured variables, not just pass/fail. Neck OD, crimp height, septum thickness, and the mean and range for each. Ten data points per dimension per lot is enough to spot a drift even if every piece is in tolerance.
• Functional results. Puncture count and leak result, pull-test result, weight-loss result with the actual mass change in mg.
• Inspector, date, and instrument IDs. If the gauge was out of calibration, the record is void and you need to know that before you accept the lot.
• Disposition. Accepted, rejected, or accepted with deviation. A deviation needs a written justification and a signature above the inspector level.

If your supplier cannot produce a record with those seven elements, the AQL number on the certificate is decoration. I have walked into receiving docks where the certificate said AQL 0.65 and the actual sample was 20 pieces pulled by the warehouse lead. The number was real. The plan behind it was not.

For buyers running a supplier qualification, the record is also the audit trail. A customer of ours, a contract lab running EPA SW-846 methods, needed to show their 2ml vial supplier had an outgoing plan that covered the attributes in their method blanks. The record with the seven elements above closed the audit in one meeting. The supplier they left had a one-page certificate and no puncture data.

Two field cases where the sampling plan was the fix

Case 1: intermittent autosampler errors traced to crimp height

A pharma QC lab was getting roughly one “no vial detected” error per 400 injections on an autosampler. The vial handler grips the crimp ring, and if the crimp height runs long the gripper does not close fully. The lab had been receiving vials with a certificate that said “crimp: pass.” We pulled 50 vials from three lots and measured crimp height with a micrometer. Two lots ran a mean of 6.02 mm with a range of 5.94 to 6.09. The third ran 5.71 mm mean. The spec was 5.5 to 5.9. The failing lots were out of tolerance on the high side, and the supplier’s “pass” was a visual check that the cap was on.

The fix was a change to the outgoing plan: crimp height became a major attribute measured with a micrometer on 32 pieces per lot, with the mean and range recorded. Error rate dropped to under one per 5,000 injections within two shipments. The lesson is not that crimp height matters (it does). The lesson is that an attribute classified as “visual” when it is actually dimensional will pass a lot that fails in use.

Case 2: a lot rejected at the dock on a critical attribute the supplier had not sampled

A distributor received 24,000 vials and the incoming inspector found septum push-through on 3 of 50 pieces. Push-through is a critical defect: the needle pushes the septum into the vial. The supplier’s record showed a Level II AQL 1.0 plan with zero major defects. The problem was that push-through was not on their defect list at all. It had been classified under “septum appearance,” which passed because the septum looked fine before puncture.

We ran a puncture rig on 32 vials from the suspect lot: 10 punctures each with a 22G needle, then a 1.5 bar leak check. 11 of 32 vials leaked at puncture 6 or later. The septum thickness on that lot ran 0.038 mm below the low limit, which let the needle pierce through rather than compress. The lot was rejected and the supplier’s plan was rewritten to add push-through as a critical attribute with an n=32, c=0 plan. The rewritten plan rejected the next lot from the same cavity set, which is exactly what a plan is supposed to do.

Skip-lot and reduced inspection: when you can trust the history

Once a supplier has 10 consecutive accepted lots under normal inspection, ANSI/ASQ Z1.4 allows reduced inspection: sample size drops to about 40% of normal, and the accept number tightens. For a 2 mL vial line running 20 lots a month, that is real inspector time saved.

I would not go to reduced inspection on the critical attributes. Ever. The critical plan stays at n=32, c=0, full stop, because the whole point of a critical attribute is that its failure mode is catastrophic and its base rate is low enough that reduced sampling cannot see it. Reduced inspection is for the minor cosmetic attributes and, with caution, the major dimensional ones. If you reduce the major dimensional plan, keep the variables data (mean and range) at full frequency even when the pass/fail count is reduced. The trend is what catches drift before it becomes a rejection.

Skip-lot is a different tool and I am more cautious about it. Skipping every other lot is fine if the process is in control and you have cavity-level traceability. It is not fine if the supplier runs multiple cavities and you have never seen a cavity-by-cavity defect map. Ask for the map before you skip anything. A supplier who cannot produce one is telling you they do not know which cavity is which.

The gauge and the lighting are part of the plan

Two operational details that get left out of the written plan and then cause arguments.

First, the lighting. “Visual inspection” without a defined illuminance is not a method. For 2 mL vials I specify 3000 lux minimum at the inspection surface, a matte gray background, and a 30 cm viewing distance, with a backlit station for particulate. Under office lighting at 500 lux, a 150 µm particle in the cavity is invisible. Under 3000 lux backlit, it is obvious. The same lot will pass or fail depending on which lamp is on, and that is a plan defect, not an inspector defect.

Second, the gauges. A go/no-go ring gauge for neck OD needs to be checked against a setting master at the start of each shift. A micrometer used for crimp height needs its zero checked daily. If the outgoing record does not name the gauge and its calibration due date, you cannot tell whether a “pass” means the part was good or the gauge was tired. I have seen a ring gauge worn 0.04 mm oversize pass a lot that a fresh gauge rejected. The gauge was the defect.

For teams building out the receiving side of this, the HPLC consumables supply chain usually has a calibration and reference-standard program you can borrow from. The same discipline applies. And if you are sourcing across multiple suppliers, a chromatography supplier with a documented outgoing plan is worth a premium over one without, because the plan is what you are actually buying when you buy the vial.

A decision checklist for the receiving dock

If you are writing or revising a plan, run these in order. Each one is a yes/no that changes the next step.

• Is every attribute on the certificate tied to a measurement method, not a visual impression? If no, stop and rewrite the defect list first.
• Are the critical attributes on a c=0 plan at n=32 or better, with a functional test (puncture, pull, leak)? If no, the AQL number on the major attributes is doing work it cannot do.
• Does the record show the drawn sample size, the defect tally by class, and the accept/reject numbers? If no, the record is a claim, not evidence.
• Are the variables data (mean and range) present for the dimensional attributes? If no, you cannot see drift before rejection.
• Is the lighting and gauge calibration defined and recorded? If no, the pass/fail is not reproducible.
• Has the supplier disclosed any resampling or deviation? If no, ask. Undisclosed resampling is the most common integrity gap.
• Is there cavity-level traceability? If no, skip-lot and reduced inspection are off the table.

For reference material on the methods side, the U.S. EPA SW-846 test methods define what the vial has to survive in environmental work, and the container requirements in those methods are a useful cross-check on your functional limits. Instrument vendors publish their own consumable specs too, and it is worth comparing your plan against what Agilent and Thermo Fisher specify for their autosamplers, because a vial that passes your plan but jams their handler is still a failed lot. Restek also publishes application notes that show the failure signatures (ghost peaks, area RSD drift) you can use to back-trace a lot to a defect class. If you are buying OEM or private-label vials, the OEM vial supplier should be able to hand you the same seven-element record under your label, and the plan should be identical to the one on their branded product.

The decision rule is this: the AQL number is the least important line on the outgoing inspection record. What decides whether a lot is releasable is whether the critical attributes were tested functionally at a sample size large enough to see the failure, whether the defect classification matches the physics of how the vial actually fails, and whether the record shows the drawn sample, the tally, and the variables data. If all three are true, an AQL 1.0 plan at Level II is fine. If any one is false, a tighter AQL will not save you, and the next 18,000 vials will sit on the dock while you argue about it.

References & further reading

Related resources from our lab consumables network:

• 2Ml Vial at 2mlsamplevial.com
• Hplc Consumables at hplcconsumable.com
• Chromatography Supplier at chromatographysupplier.com
• Oem Vial at oemsamplevials.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