A lot of 3,000 11 mm grey butyl septa left the moulding floor on a Tuesday and reached the customer’s incoming dock on Friday with a Certificate of Conformance that said “meets specification.” By the following Wednesday, the customer’s QC group had rejected the lot: 14 of 200 sampled closures leaked at 1.5 bar helium, and 9 showed visible flash across the needle path. The supplier had no septa AQL sampling plan at all. They had a visual check by the line operator and a signed piece of paper. That is the gap this article is about.
Outgoing QC on septa is not the same as incoming inspection, and it is not the same as in-process capability. This is the last gate before the pallet leaves your building. It answers one question: does this lot, as shipped, meet the agreed acceptance criteria at a defined confidence level? If you cannot answer that with a sample size, an AQL number, and a defect classification, you do not have outgoing QC. You have a signature.
The single most common failure I see in OEM septum contracts is a sampling plan reverse-engineered from whatever the operator can comfortably inspect in a shift. A 32-piece sample gets pulled because 32 fits on a tray. Nobody asks what AQL that sample actually demonstrates. Then the buyer assumes they are getting 0.65 and they are getting something closer to 4.0.
For septa, I recommend splitting the plan across three defect classes. This is the ISO 2859-1 framework applied to a product where the failure modes are physical, not dimensional alone.
| Defect class | Typical defects | AQL (inspection level II) | Lot size 1,201 to 3,200 |
|---|---|---|---|
| Critical | Through-hole, missing PTFE facing, wrong polymer, contamination visible at 10x | 0.065 | n = 125, Ac = 0 |
| Major | Flash on needle path, off-centre disc, delamination of facing, hardness out of band | 0.65 | n = 125, Ac = 2 |
| Minor | Cosmetic surface marks, mould parting line visible, colour drift within tolerance | 2.5 | n = 125, Ac = 7 |
Note that the sample size is the same across all three classes. That is deliberate. You pull one sample of 125 and inspect it against three criteria. Pulling three separate samples wastes operator time and creates confusion about which piece belongs to which plan. The Ac (acceptance number) changes, not the n.
For a lot of 501 to 1,200 pieces, n drops to 80. For 151 to 500, n = 50. For 91 to 150, n = 32. For 26 to 90, n = 20. Below 26 pieces, you inspect 100%. This is straight from the ISO 2859-1 general inspection level II table and it is what I would put in any OEM supply agreement. If a supplier pushes back and wants to use level I (smaller samples), ask them to justify it against the critical defect class. They usually cannot.
Two more decisions belong in the plan itself, not in a side conversation. First, the sampling method. Draw pieces from at least three positions in the lot: the top carton, a middle carton, and the bottom carton, and within each carton pull from at least two tray positions. A 125-piece sample pulled entirely from the top carton of a 3,000-piece lot is not a random sample, it is a sample of the top carton. Second, the retest rule. If a lot fails on the major class at Ac = 2 with 3 defects found, the lot is rejected. No retest, no “pull another sample and see.” Retesting to acceptance is how a supplier converts a 0.65 AQL into a 4.0 in practice, and I have watched it happen on more than one incoming dock.
When to go to tightened inspection
ISO 2859-1 switching rules matter here. If two out of five consecutive lots are rejected on the same defect class, you switch to tightened inspection. That means the same sample size but a lower Ac. For the critical class at 0.065, tightened drops Ac to 0 and effectively doubles the effective sampling severity because the acceptance number cannot go below zero. For the major class, Ac goes from 2 to 1.
I have seen suppliers quietly stay on normal inspection after three rejections because nobody owned the switching rule. Put it in the contract. Write the trigger into the outgoing inspection record template so the operator cannot skip it.
The reverse switch matters too, and it is where honest suppliers get punished by sloppy buyers. After five consecutive lots accepted on tightened inspection, you return to normal. If the buyer’s ERP does not track the switch-back, the supplier stays on tightened for eighteen months and starts pricing the extra inspection into the quote. The switching rule has to work in both directions or it is just a penalty.
What a realistic outgoing inspection record actually contains
Most CoAs I audit are one page and contain almost nothing useful. A real outgoing inspection record for a septum lot has eight fields minimum, and I want all eight on the same document that ships with the pallet.
- Lot/batch identifier traceable to the moulding date, cavity number if the tool is multi-cavity, and polymer grade lot.
- Sample size and inspection level, with the standard cited (ISO 2859-1, level II, normal or tightened).
- Defect classification results by class, with the actual count found against the Ac number.
- Dimensional results: disc diameter (typically 11.0 ± 0.2 mm for a 11 mm closure), thickness (commonly 1.5 ± 0.15 mm for butyl, 0.9 to 1.0 mm for silicone), and centre hole diameter if applicable. Report the mean and range, not just pass/fail.
- Physical test results: hardness (Shore A, typically 50 to 60 for butyl, 45 to 55 for silicone), and a puncture/reseal count on a defined needle gauge.
- Functional test: leak rate at a defined pressure (I use 1.0 bar helium for 30 seconds, or a dye ingress test at 0.5 bar for 10 minutes).
- Operator ID and date/time of inspection. Not the shift supervisor’s name. The person who actually did the inspection.
- Disposition: accept, reject, or accept under deviation with the deviation number referenced.
If the CoA you receive does not have at least the sample size and the defect counts, it is a marketing document. Ask for the raw inspection sheet. A supplier who cannot produce it does not have an outgoing QC process, they have a stamp.
The puncture/reseal count deserves its own line. For a standard 11 mm butyl septum used on an autosampler vial, I want to see a reseal test at 5 punctures with a 23-gauge needle, leak-checked after each. A silicone/PTFE sandwich should hold 10 to 15 punctures before the facing shows a permanent track. If the CoA reports “puncture test: pass” with no number, that is not a result.
The measurements that belong on the record and the ones that do not
There is a temptation to put every instrument reading on the CoA. Resist it. The record should carry the measurements that discriminate between a good lot and a bad one, and nothing else. On a butyl septum, that is hardness, thickness, disc diameter, compression set after 22 hours at 100 °C (target under 25% for a 50 to 60 Shore A compound), and the puncture/reseal count. On a silicone/PTFE sandwich, it is facing thickness (typically 0.13 to 0.20 mm), facing adhesion (a 180° peel test, target above 0.5 N/cm), and the same puncture/reseal count. Colour, gloss, and mould parting line appearance belong on the minor class and nowhere near a numeric column.
One number I insist on that suppliers rarely volunteer: the compression set. A septum that loses 30% of its thickness under autosampler cap torque will leak on the second injection even if it passed the first. Compression set is the measurement that predicts field failure, and it takes 22 hours, so it cannot be run on every lot in real time. Run it on the first lot of each polymer grade lot and reference the result on every subsequent CoA from that grade lot. That is honest and it is fast.
Three field cases where the sampling plan caught what visual inspection missed
Case 1: the hardness drift that only showed at puncture 4
Symptom: A customer running EPA Method 8260 volatiles on a HPLC vial platform reported increasing needle coring after about 6 hours of autosampler runtime. The septa looked fine on the bench. Visual inspection passed every lot for three months.
Measurement: We pulled the retention samples and ran Shore A hardness across 30 pieces per lot. The suspect lot averaged 62 Shore A against a spec of 50 to 60. The prior three lots averaged 54. A 8-point hardness shift on butyl translates to a measurable increase in needle drag and a higher coring rate. The SW-846 method does not specify septum hardness, which is exactly why the supplier’s incoming spec sheet never included it.
Fix: Hardness became a major-class defect in the AQL plan, sampled at n = 125, with a durometer reading on 10 pieces per sample. The cure-time drift on the press was traced to a thermocouple that had been reading 4 °C low for six weeks. The outgoing inspection caught it in one lot because the plan required a measured value, not a visual.
Case 2: the delaminated PTFE facing that passed leak test
Symptom: A pharma OEM customer rejected a 2,400-piece lot of silicone/PTFE vial septa after 11 vials showed ghost peaks in blank injections. The septa had passed the supplier’s leak test at 0.5 bar.
Measurement: We cross-sectioned 20 pieces from the retained sample and found PTFE facing delamination on 4 of them, all from cavity 3 of a 16-cavity tool. The leak test passed because the delamination was on the underside of the facing, not across the seal path. The defect was invisible to the leak test and to the eye. It only appeared under a 10x loupe with side lighting.
Fix: The outgoing plan added a 10x visual inspection for facing delamination as a critical defect, Ac = 0. Cavity 3 was pulled for tool maintenance. The lot was scrapped. The real lesson: a leak test alone does not catch facing defects. You need a visual at magnification as a separate inspection step, and it has to be in the sampling plan, not left to the operator’s discretion.
Case 3: the off-centre disc that passed dimensional but failed the needle guide
Symptom: An OEM filling line running 30 vials a minute started bending needles on one autosampler position. The septum supplier’s dimensional report showed disc diameter at 11.02 mm mean, well inside the 11.0 ± 0.2 tolerance. Nothing on the CoA looked wrong.
Measurement: We measured concentricity, not diameter. The centre hole on 7 of 40 sampled pieces was displaced by 0.4 to 0.6 mm from the geometric centre. Diameter was fine. Position was not. The needle guide on that autosampler position had a 0.3 mm clearance, so a 0.5 mm displacement pushed the needle into the guide wall on every stroke.
Fix: Concentricity went into the major class with a 0.25 mm tolerance, measured on 10 pieces per sample with a vision system. The root cause was a worn core pin in the mould, which the supplier’s dimensional check never looked at because the spec sheet only listed diameter. A diameter-only sampling plan cannot catch a position defect, and position is what the needle guide cares about.
Sample size math that procurement will actually accept
Procurement teams push back on n = 125 because it feels like a lot of pieces to pull from a 2,000-piece lot. It is 6.25%. The inspection takes about 25 minutes for a trained operator with a loupe, a durometer, and a leak fixture. That is a rounding error against the cost of a rejected shipment.
Here is the argument that closes the conversation. At n = 125 with Ac = 0, you demonstrate a defect rate of approximately 2.4% at 95% confidence. That is the upper bound, not the actual rate. If you want to demonstrate 1.0% at 95% confidence with Ac = 0, you need n = 299. If you want 0.65%, you need n = 460. The AQL number and the sample size are locked together by the binomial distribution, and no amount of supplier confidence changes the math.
So when a supplier says “we sample 30 pieces and we have never had a problem,” what they are actually demonstrating is that their defect rate is probably below 10%, maybe. That is not a quality claim. That is a coin flip with extra steps.
For septa specifically, I would add one wrinkle. The defect rate is not uniform across the lot. Cavity-to-cavity variation on a multi-cavity tool can be 3x to 5x on flash and facing defects. A random sample of 125 from a 16-cavity tool will, on average, hit each cavity about 8 times. That is enough to catch a bad cavity, but only if you record the cavity number on each piece. If your supplier cannot tell you which cavity a septum came from, the AQL plan is sampling a blend, and a single bad cavity can hide behind 15 good ones.
The cost side is worth stating in numbers a buyer will recognize. A 125-piece sample at 25 minutes of operator time, loaded at a fully burdened rate of $38 an hour, is $15.83 per lot. Add the leak fixture consumables (helium at roughly $0.40 per lot) and the durometer wear (negligible) and you are at about $16.50 per lot inspected. A single rejected shipment to a pharma customer, with the return freight, the deviation report, the customer’s line downtime, and the corrective action meeting, runs between $4,000 and $12,000 depending on how the customer charges downtime. The inspection pays for itself on the first escape it prevents, and it prevents roughly one escape in every 40 lots at a 1% true defect rate. Do that arithmetic in the meeting and the sample size argument ends.
Defect classification is where most plans quietly fall apart
The table above is only useful if the operator can classify a piece consistently. That requires a defect catalogue with photographs, and it requires that the catalogue be agreed with the customer in writing. I have watched two competent inspectors disagree on the same piece because one called a 0.3 mm flash on the needle path a major and the other called it a minor. That disagreement is a sampling plan failure even though the sample size was correct.
My rule is mechanical, not judgmental. Flash on the needle path is a major defect if it exceeds 0.15 mm in height or extends more than 0.5 mm along the path. Below that, it is minor. Facing delamination of any visible extent is critical. A disc that is off-centre by more than 0.25 mm is major. A colour deviation that exceeds a ΔE of 2.0 against the master is minor. Write those thresholds into the catalogue with a photograph of a borderline piece at each threshold. The inspector compares, not decides.
The catalogue also settles the argument about cosmetic defects. Buyers who put every surface mark into the major class end up rejecting lots that are functionally perfect, and suppliers who put every surface mark into the minor class ship lots that look bad on a customer’s line. The ΔE 2.0 threshold and a 0.15 mm flash limit are defensible to both sides. Everything else is negotiation.
The decision rule I would put in every OEM supply agreement
Strip out everything else and this is the clause that matters:
Every lot ships with an outgoing inspection record citing ISO 2859-1 level II, normal inspection, with critical AQL 0.065 (Ac = 0), major AQL 0.65 (Ac = 2), and minor AQL 2.5 (Ac = 7) at n = 125 for lots of 1,201 to 3,200 pieces. Tightened inspection applies after two rejections in five consecutive lots. Records include sample size, defect counts by class, hardness mean and range, dimensional mean and range, concentricity, leak test pressure and duration, puncture/reseal count, and operator ID.
If a supplier cannot sign that clause, they are telling you something about their process. Believe them.
The related question of whether the vendor actually measured anything at all is covered elsewhere; here the point is narrower. The outgoing record is the last verifiable data point before the pallet leaves. If it does not contain a sample size and a defect count, it is not data. It is a hope.
One more thing worth saying plainly. The AQL plan does not make the product good. It makes the escape rate predictable. A supplier with a 3% true defect rate and a proper AQL plan will reject lots regularly and honestly. A supplier with a 0.3% true defect rate and no plan will ship bad lots occasionally and never know why. The plan is how you find out which supplier you actually have. For the full range of septum formats that go through this gate, see Septa. For reference methods on the analytical side that the septa must survive, the Restek technical library, Agilent’s GC consumables documentation, and Thermo Fisher’s chromatography resources are the places I send new QC hires first.
Set the AQL at 0.065 for critical defects, sample 125 pieces per lot at level II, record the counts, and switch to tightened after two rejections in five lots. If the outgoing record cannot show those three things, the lot does not ship.
References & further reading
Related resources from our lab consumables network:
- Septa at septacaps.com
- Vial Septa at vialsepta.com
- Hplc Vial at hplcgc.com
- Autosampler Vial at autosamplerhplc.com
External references and standards:
