Product manufacturing uses a wide range of processing centers

The outgoing inspection sheet for lot 24-117 showed 1,600 pieces sampled from 40,000, six majors, zero criticals. Eleven days later a fill line in Raleigh rejected the lot: 14 of 60 stoppers on the first tray pulled a vacuum leak at 30 inHg. The QC record had passed because the sampling plan read AQL 1.0, Level II, n=200 from a table someone copied out of a 1994 supplier manual, and the inspector had counted 200 closures, not 1,600. The plan on paper and the plan in practice were two different documents. That gap is what a vial closures AQL sampling plan and outgoing QC record has to close before anything leaves the dock.

An AQL of 1.0 on a 40,000-piece lot means you are sampling 200 pieces, not 1,600

Start with the arithmetic, because most of the arguments I have had about sampling plans were really arguments about table lookup. Under ISO 2859-1 (the general inspection standard most closure suppliers quietly follow, whether they cite it or not), a lot of 35,001 to 150,000 units at General Inspection Level II gives code letter M. Code letter M at AQL 1.0 gives a sample size of n=315 with an acceptance number of 5 and a rejection number of 6. At AQL 0.65, the same code letter M gives n=315, Ac=3, Re=4. At AQL 0.25, n=315, Ac=1, Re=2.

Notice what does not change across those three rows: the sample size. What changes is the acceptance number. That is the single most misunderstood thing about AQL plans in this industry. Tightening the AQL does not mean inspecting more pieces. It means failing the lot on fewer defectives. If a supplier tells you they “moved to AQL 0.65” without telling you the sample size, ask them what the Ac/Re numbers are. If they cannot answer in ten seconds, the plan is decorative.

For vial closures specifically, I use a tighter structure than the generic tables because the defect classes are not equally damaging:

Those numbers are not sacred. They are defensible. The point is that each class gets its own sample size and its own acceptance number, drawn from the same lot, and the inspector must know which class a defect falls into before counting it. If your plan has one AQL for “vial closures,” you do not have a plan. You have a lottery ticket with a form number on it.

Defect classification is where the plan actually lives or dies

The AQL table is easy. The classification scheme is hard, and it is where I see suppliers and buyers diverge most often. A flash ring on the sealing land of a 20 mm butyl stopper is a Major A because it prevents the stopper from seating flush against the vial crown, and a 0.2 mm gap at 30 inHg will leak. The same flash on the outside of the cap skirt is Minor. The inspector needs a photo reference card with a scale bar, not a verbal description. I have seen two inspectors classify the same defect differently on the same lot because the work instruction said “excessive flash” and never defined “excessive.”

Three classification rules I insist on:

• Any defect that touches the sealing interface is at least Major A. That means the septum face, the stopper plug, the cap sealing ring, and the crown contact surface. No exceptions, no “it’s only 0.1 mm.”
• Dimension out of tolerance is a defect, not a variable. If the plug diameter is specified at 7.65 ± 0.05 mm and you measure 7.72 mm, that is a Major A, even if it “looks fine.” Variable data belongs in the capability study, not the attribute inspection. (Capability and incoming inspection belong to a separate discussion; here we only care what the outgoing record says.)
• Particulate is critical, always. A visible particle on the septum face of a closure destined for an HPLC vial is a critical defect regardless of size. I have seen a single 80 µm silicone shard on a septum face cause a 3.2% RSD spike on a 6-injection sequence in a validated method. That is a critical defect, not a cosmetic one.

Write the classification card before you write the sampling plan. If the card has more than 12 defect codes, it is too long for a shop-floor inspector to apply consistently at line speed. Twelve codes, photo references, and a scale bar. That is what works.

What a realistic outgoing inspection record contains (and what most of them are missing)

The record is the product of the inspection, not a byproduct. If a customer audit cannot reconstruct the lot from the record alone, the record is incomplete. Here is what I require on every outgoing certificate, and I have walked enough audits to know which fields get skipped:

• Lot number and quantity. Both the molding lot and the assembly lot if they differ. Closures are frequently molded in one run and assembled in another, and a defect in the assembly lot cannot be traced to molding if the record only carries one number.
• Sampling plan reference. The standard (ISO 2859-1), the inspection level (II), the AQL per defect class, the sample size, and the Ac/Re numbers. All five, on the record, not in a procedure document filed somewhere.
• Sample size actually inspected. Not the planned sample size. The actual count. If the inspector stopped at 200 because the shift ended, the record must say 200.
• Defect count by class. Criticals, Major A, Major B, Minors, each with a count and a disposition (accept, reject, rework, sort).
• Measurement data for the critical dimensions. Plug diameter, flange thickness, cap height, and where applicable the crown-to-flange compression dimension. For a 20 mm butyl stopper, I want plug diameter at 7.65 ± 0.05 mm, flange thickness at 1.5 ± 0.1 mm, and a hardness reading at 50 ± 5 Shore A. Report the mean and the range, not just “within spec.”
• Environmental conditions. Temperature and RH at the time of inspection. Elastomer hardness shifts measurably between 18 °C and 28 °C, and a hardness reading taken at 30 °C on a butyl stopper will read 2 to 3 Shore A points low. If the record does not state the conditions, the hardness number is not defensible.
• Inspector ID, date, and equipment ID. The durometer, the optical comparator, the balance. Calibration status of each, referenced to a calibration ID.
• Disposition and release signature. Who released the lot, and against what criteria.

What most records are missing: the actual sample size, the environmental conditions, and the equipment IDs. Those three omissions are what turn a passing record into a failed audit. I have seen a supplier lose a $400k annual contract over a missing durometer calibration ID on a certificate, not over a defective lot.

Two field cases where the plan failed before the product did

Case 1: The 0.3 mm flash that passed AQL 1.5 and leaked at 30 inHg

Symptom: A customer running 20 mm chlorobutyl stoppers on a 10 mL fill reported 4 to 6 leak rejects per 1,000 vials on a vacuum decay test at 30 inHg, holding for 5 seconds. The outgoing record showed zero majors.

Measurement that identified it: We pulled 315 stoppers from the retained sample and measured flash on the sealing land under a 10× loupe with a reticle. Thirteen pieces had flash between 0.15 mm and 0.35 mm. The outgoing plan classified flash as Minor at AQL 4.0, and the Ac was 10. Thirteen defectives against Ac=10 should have failed the lot. The inspector had counted flash as “present or absent,” not measured it, and counted 6 pieces as having flash. The classification card had no dimension threshold.

Fix: Flash on the sealing land was reclassified to Major A at AQL 0.65 with a hard threshold of 0.10 mm measured with a reticle. The Ac/Re became 3/4. The next three lots failed outgoing inspection at 4, 5, and 4 defectives respectively, and the molding tool was pulled for a flash land rework. Leak rejects dropped to under 0.5 per 1,000 within two lots.

Case 2: The hardness drift that no one measured because the record said “within spec”

Symptom: A CRO running a stability study on a biologic saw coring increase from 1.2 to 3.8 fragments per 100 punctures on a 13 mm stopper over a six-month period. The stoppers were from the same nominal lot family.

Measurement that identified it: We pulled retained samples from three outgoing lots and measured hardness at 23 °C and 50% RH, 10 readings per lot, on a Shore A durometer with a 1 kg load. Lot A read 50.2 ± 0.8, Lot B read 53.6 ± 1.1, Lot C read 56.1 ± 0.9. All three were “within spec” against a 50 ± 5 Shore A specification. The outgoing records for all three said “hardness: pass” with no number. The coring increase tracked the hardness increase at roughly 0.4 additional fragments per 100 punctures per Shore A point.

Fix: The outgoing record now carries the mean and range for each lot, and the specification was tightened to 50 ± 3 Shore A for the specific product family. The supplier added a 24-hour post-cure conditioning step before outgoing inspection to reduce lot-to-lot hardness spread. Coring returned to 1.4 fragments per 100 punctures. The specification change was the fix; the record change was what made the fix visible.

Neither case was a sampling plan failure in the arithmetic sense. Both were classification and recording failures. The AQL table was correct in both cases. The record was not.

Sampling frequency, not just sample size, is where the plan earns its keep

AQL plans tell you how many pieces to inspect from a lot. They do not tell you how many lots to inspect, or when to switch to tightened or reduced inspection. That is the part most suppliers leave out, and it is the part that catches drift before it becomes a rejected shipment.

My default frequency rules for vial closures:

• Normal inspection on every lot, using the table above. No skipping.
• Skip-lot only after 10 consecutive lots at normal inspection with zero criticals and defect rates below half the AQL. Even then, skip no more than every other lot, and never skip the first lot after a tool change or a material lot change.
• Tightened inspection (switch to the next tighter AQL row, or increase sample size by one code letter) after 2 lots in 5 rejected at normal. Return to normal only after 5 consecutive accepted lots.
• Reduced inspection is allowed under ISO 2859-1 but I do not use it for closures. The savings are trivial and the risk is not. A reduced plan on a 40,000-piece lot saves maybe 90 pieces of inspection time. It costs you the ability to catch a 0.3 mm flash trend.

The frequency rules belong on the outgoing record too, not just in the procedure. If the record does not state whether the lot was inspected at normal, tightened, or reduced, the customer cannot tell whether a passing result means what they think it means.

The receiving-side check: what to do with the record when it arrives

If you are the buyer, the outgoing record is an input to your own decision, not a substitute for it. Three things to check before you release a lot to production:

1. Does the sample size match the lot size and the AQL? If the lot is 40,000 and the record shows n=125, the supplier is using a reduced plan or a Level I plan. Ask which, and why.
2. Are the defect classes defined, or is there a single “defects” count? A single count means the supplier has not classified. That is a red flag on the process, not just the record.
3. Are the critical dimensions reported as numbers, or as “pass”? “Pass” is not data. Mean and range, or it did not happen.

For the dimensional side of the receiving check, a quick go/no-go on the plug diameter and flange thickness with a calibrated pin gauge and a 0.001 mm micrometer takes about 90 seconds per lot and catches the drift that the supplier’s attribute inspection will miss. The vial closures you buy are only as good as the dimensional data that travels with them.

A practical outgoing inspection sequence you can run in 45 minutes per lot

Here is the sequence I have run on thousands of lots. It fits in 45 minutes for a 40,000-piece lot at Level II:

1. Pull the sample (5 min). Random from at least 8 cartons across the pallet, not from the top carton. Use a random number table or a simple spreadsheet RAND between 1 and carton count. The top carton is not random, and the top carton is where the flash from the last cavity tends to sit.
2. Visual inspection (15 min). 315 pieces under 2× magnification with a 0.5 mm reticle. Count criticals first, then Major A, then Major B, then Minors. Stop counting a class once the Re number is exceeded and record the count as “≥Re.”
3. Dimensional check (10 min). 10 pieces from the sample, plug diameter and flange thickness, mean and range. If the range exceeds 60% of the tolerance band, escalate to the process engineer even if the mean is centered.
4. Hardness (5 min). 5 pieces, 3 readings each, at 23 ± 2 °C. Record mean and range. If the shop floor is above 26 °C, move the sample to a conditioned room for 30 minutes first.
5. Functional check (10 min). 10 closures assembled onto the specified vial, crimped or screwed to the specified torque (for crimp, 1.2 to 1.5 N·m on a 20 mm aluminum seal; for screw, 0.8 to 1.0 N·m on a 13 mm cap), and leak-tested at 30 inHg for 5 seconds. Any leak is a critical.
6. Record and release (5 min). Fill the record, attach the photos of any defect found, sign, release or hold.

That sequence is not exhaustive. It is repeatable. Repeatability is what an audit checks, and repeatability is what catches drift. The analytical testing lab on the receiving end will catch what the outgoing plan misses, but by then the lot is in your warehouse and the freight is sunk.

Where the standards actually help, and where they do not

ISO 2859-1 gives you the tables. It does not give you the defect classification, the frequency rules, or the record format. Those are yours to define, and they are what a customer audit will actually examine. EPA SW-846 and the associated data quality objectives are useful as a model for how to write acceptance criteria that a third party can verify, even though they were written for environmental sampling, not closures. The DQO framework (state the decision, state the uncertainty, state the sampling design) maps cleanly onto an outgoing inspection plan.

For the chromatographic side, the consumables suppliers publish useful reference material on septa and closure performance that is worth reading before you set your acceptance numbers. Agilent’s GC columns and accessories documentation and Restek’s technical resources both discuss septa bleed, coring, and injection port temperature limits in enough detail to inform a defect classification card. Thermo Fisher’s chromatography and sample handling pages cover the same ground from the instrument side. None of them will give you an AQL table for closures. That is your job.

The chromatography consumables you source are only as consistent as the outgoing plan behind them. A supplier who cannot show you the Ac/Re numbers on the certificate is a supplier who has not thought about the plan.

The decision rule

If a supplier cannot produce, on request and within one business day, a lot-specific outgoing record that states the standard, the inspection level, the AQL per defect class, the actual sample size inspected, the defect count by class, and the measured values for the critical dimensions, do not release the lot. Not because the closures are necessarily bad, but because you cannot tell whether they are good. The AQL table is a commodity; every supplier has it. The classification card and the record are the differentiator, and they are the only part of the plan that a customer audit will actually score. Set your AQL at 0.65 for anything that touches the sealing interface, sample at Level II, and write the record as if the auditor is reading it cold. That is the whole plan.

References & further reading

Related resources from our lab consumables network:

• Vial Closures at vialsclosures.com
• Analytical Testing at testinganalytical.com
• Hplc Vial at vialforhplc.com
• Chromatography Consumables at chromatographyconsumables.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