What an AQL sampling plan actually commits you to

On a Tuesday afternoon in March, a shipment of 200,000 septa left our dock and arrived at a customer’s loading bay four days later. The receiving QC tech pulled a single box, ran a puncture test on 10 septa, and found two that coring-failed at 40 punctures instead of the specified 80. She rejected the entire lot. She was right to flag it, but the sampling plan she used (10 pieces from one box) had a 94% chance of missing the same defect rate in the other 199 boxes. That is the problem with outgoing QC on chromatography products: the AQL sampling plan is the only thing standing between a defect in the factory and a failed run in someone else’s lab, and most small and mid-size suppliers don’t have one worth the paper it’s printed on. This article is about how to build one that holds up, what AQL levels actually mean for septa, caps, ferrules, and vials, and what an outgoing inspection record should contain.

ANSI/ASQ Z1.4 (the U.S. adoption of ISO 2859-1) is the reference most chromatography manufacturers work from, and it is a contract with the customer whether anyone on the sales side realizes it or not. When you pick an AQL of 1.0 at inspection level II for a lot of 10,000 pieces, you are saying: I will sample 200 pieces, and I will accept the lot if I find 5 or fewer defectives. If I find 6, I reject. That is the whole mechanism. The number 200 and the number 5 are not opinions; they come out of a table indexed by lot size, inspection level, and AQL.

Two things trip people up. First, the AQL is not the defect rate you ship. It is the defect rate you are willing to accept 95% of the time. A 1.0 AQL plan accepts lots that are actually running at 1.5% defective more often than most quality managers want to admit. I have watched a supplier claim “1.0 AQL” for years while shipping lots that were 2 to 3% defective on the coring attribute, because the plan was 1.0 AQL and the process was running at 2.5%, and the samples kept passing. The plan was working. The process was not.

Second, inspection level matters as much as AQL. Level II is the default. Level I halves the sample size and doubles your risk. Level III doubles the sample and is what I would insist on for any attribute where a single failure takes down a customer’s run: needle coring, cap seal integrity, ferrule ID tolerance. For a lot of 10,000 septa at level III and 0.65 AQL, you’re sampling 315 pieces and accepting on 3. That is a very different commitment from 200 and 5.

Choosing the AQL by defect class, not by product

The mistake I see most often is a single AQL applied to every attribute on a part. That is wrong. A ferrule with a scratched exterior is cosmetic. A ferrule with an undersized ID is a leak. Those two attributes cannot share a plan.

Classify defects first, then assign AQLs:

  • Critical (AQL 0, zero acceptance): anything that causes a safety issue or a guaranteed analytical failure. Wrong septum thickness that prevents cap seating, a ferrule with the wrong taper angle, a vial with a visible crack at the neck. Sample size is whatever the plan says, but the acceptance number is always zero. One finding, lot rejected.
  • Major (AQL 0.65 to 1.0): functional attributes that will cause a customer complaint or a failed run if they reach the bench. Septum coring at the rated puncture count, cap torque out of band, ferrule ID out of tolerance, septum durometer outside spec, particulate contamination above the limit.
  • Minor (AQL 2.5 to 4.0): cosmetic and handling attributes. Scratches, label misalignment, minor color variation on caps, packaging dents that don’t compromise the seal.

For chromatography consumables specifically, I would run critical at zero acceptance, major at 0.65 for anything that touches the sample path, and 1.0 for physical attributes like dimensions that a customer can inspect on receipt. Minor at 2.5. That is tighter than the general-purpose consumer goods default and it should be, because the cost of a failed injection is not a returned vial, it is a 45-minute rerun and a missed turnaround.

Sample sizes that match the lot, not the habit

The most common bad habit is a fixed sample size. “We pull 20 pieces from every lot.” That is fine for a lot of 500 and meaningless for a lot of 200,000. Z1.4 ties sample size to lot size, and the curve flattens above about 35,000 pieces, but it does not flatten to nothing. Here is the working table I keep on the wall, drawn from Z1.4 general inspection level II:

Lot size (pieces) Sample size (level II) Accept at AQL 1.0 (major) Accept at AQL 0.65 (critical-adjacent)
151 to 280 32 1 0
281 to 500 50 1 0
501 to 1,200 80 2 1
1,201 to 3,200 125 3 1
3,201 to 10,000 200 5 2
10,001 to 35,000 315 7 3
35,001 to 150,000 500 10 5
150,001 to 500,000 800 14 7

Two operational notes. First, the sample must be pulled from across the lot, not from one carton. If your lot is 40 boxes of 5,000 septa, the sample has to touch at least 10 boxes, drawn from top, middle, and bottom of the pallet. A sample of 200 from a single box is not a sample of the lot; it is a sample of that box. Second, the sample has to be random in a defensible way. I have seen inspectors pull from the front row because it’s easier. That is not random, and if a customer audits your inspection records, it will show.

When to switch to tightened or reduced inspection

Z1.4 has a switching rule most suppliers ignore. If two out of five consecutive lots are rejected on the same defect class, you switch to tightened inspection: same AQL, but the acceptance number drops by one step and the sample size increases. If ten consecutive lots are accepted under normal inspection, you can go to reduced inspection: smaller sample, same AQL. The point is that the plan responds to the process. A supplier running the same sample size and acceptance number for three years is not running a sampling plan; they are running a ritual.

Defect classification for chromatography parts, attribute by attribute

Generic defect lists do not work for this product family. Here is how I would classify the attributes that actually show up in outgoing inspection of the parts most suppliers ship: septa, caps, vials, ferrules, and liners.

Septa

  • Critical: thickness outside ±0.05 mm of nominal (a 3.0 mm septum that measures 2.8 mm will not seat under a crimp cap reliably), durometer outside the specified Shore A band by more than 3 points, visible voids or delamination.
  • Major: coring failure before the rated puncture count (this is where the 80-puncture spec lives; test at 40 and 80), extractable level above the method limit, off-odor, contamination visible under 10× magnification.
  • Minor: surface blemishes that don’t affect sealing, edge flash, slight color variation between batches of the same silicone.

The puncture test is the one that matters. Run it on a fresh needle, at the specified gauge, at a controlled rate, and count coring events (visible plug or fragment). Do not run it on a bench needle that has already been used 200 times; you will get false failures. I use a dedicated 22-gauge needle and replace it every 500 punctures.

Caps and closures

  • Critical: cap does not seat to the specified torque (for screw caps, 0.5 to 1.2 N·m depending on the thread and septa thickness), visible crack at the skirt, wrong thread pitch (GPI 13-425 vs. 8-425 confusion is common and catastrophic).
  • Major: septa retention failure (septum falls out when cap is inverted), torque out of band, dimensional mismatch with the specified vial neck OD.
  • Minor: color variation, mold flash, label misregistration.

GPI thread standards are worth pinning to the wall. A 13-425 cap is not interchangeable with an 8-425, and a supplier who ships the wrong one will not find out until the customer’s autosampler jams. If you want a deeper look at how caps and septa interact dimensionally, the team at Vials With Caps has done good work on the fit question.

Vials

  • Critical: neck crack, wrong neck OD, bottom thickness outside spec (affects autosampler needle depth), visible glass defects in the sample path.
  • Major: dimensional tolerance on height and OD (typically ±0.2 mm on a 12×32 mm vial), interior volume below nominal, particulate above the method limit.
  • Minor: exterior scratches, label placement, graduation mark alignment.

For 2 mL autosampler vials, the critical dimensions are 12.0 ±0.2 mm OD, 32.0 ±0.3 mm height, and 6.0 ±0.2 mm neck OD for the 11 mm crimp format. Anything outside that will cause needle misalignment on at least one of the common autosampler platforms. If you’re sourcing vials, Vial Supplier is a good starting point for comparing spec sheets against your autosampler’s requirements.

Ferrules and liners

  • Critical: ferrule ID out of tolerance (a 0.4 mm ferrule that measures 0.45 mm will leak), taper angle wrong, liner ID out of spec, deactivation failure on liners.
  • Major: ferrule hardness out of band (too soft = over-compression at the specified torque, too hard = under-sealing), liner contamination, wrong O-ring material.
  • Minor: cosmetic surface finish, packaging.

What a realistic outgoing inspection record contains

Most inspection records I see are useless. They say “passed” with a date and an initial. That is not a record; it is an assertion. A record that survives a customer audit and a supplier corrective action contains at minimum:

  • Lot ID and traceability link to the batch record (the manufacturing batch, the raw material lots, the molding or assembly date).
  • Sampling plan used: standard reference (ANSI/ASQ Z1.4 or ISO 2859-1), inspection level, AQL per defect class, sample size, acceptance number.
  • Attribute results: for each attribute, the measured value or pass/fail, the test method, the gauge or instrument ID, and the calibration due date of that instrument.
  • Defect tally: number of critical, major, and minor findings, with a description of each finding (not just “cosmetic”).
  • Disposition: accept, reject, or accept with deviation, with the deviation reason and the approver.
  • Environmental conditions at the time of inspection if any attribute is temperature- or humidity-sensitive (septum durometer, torque measurement).
  • Inspector ID and date/time.

I have seen a customer reject a lot because the inspection record showed a torque measurement of “0.8 N·m” with no gauge ID and no calibration date. The lot was probably fine. The record wasn’t. In a regulated environment, an uncalibrated measurement is not a measurement.

A field case: the septa that passed at 80 and failed at 40

We shipped a lot of 120,000 silicone/PTFE septa to a pharma customer. Incoming QC at the customer pulled 50 pieces, ran the puncture test at 80 punctures, and found zero coring. They accepted. Two weeks later, a method development team reported coring at 35 to 40 punctures on the same lot. The problem was the test protocol, not the septa. Our outgoing test had been run at 80 punctures only, which is the rated spec. The customer’s method was running 40 punctures per septum on average because their autosampler was doing a pre-pierce and a sample pierce per injection. The fix was to add a 40-puncture intermediate test to the outgoing protocol for this product line, with the same acceptance criteria. The septa were fine at 80; they were marginal at 40. We changed the spec to require coring-free performance at both 40 and 80 punctures, and we changed the sample size for that attribute to level III (315 pieces for this lot size). That is what a defect class change looks like in practice.

Two failure modes that only show up in outgoing sampling

The lot that drifts inside the acceptance band

A ferrule lot measured 0.40 ±0.02 mm ID on the incoming material cert. Our outgoing sample of 125 pieces showed a mean of 0.415 mm and a range of 0.395 to 0.432. Every piece was inside the 0.38 to 0.42 spec. Every piece passed. The lot was accepted. Two months later, a customer reported leaks on a GC inlet at 350 °C. The problem was not the mean; it was that the distribution had shifted toward the upper limit, and at 350 °C the ferrule was compressing more than it should, and the seal was failing at the specified torque. The fix was to add a mean-shift trigger to the outgoing inspection: if the sample mean is more than 1 standard deviation off the historical process mean, the lot is flagged for review even if every individual piece is in spec. That is a control chart logic applied at outgoing inspection, and it catches the drift that a pure attribute plan misses.

The cosmetic defect that was actually a functional one

A cap lot showed minor surface scratches on about 3% of the sample. Under our plan, minor defects at AQL 2.5 were acceptable, so the lot passed. The scratches turned out to be from a worn mold insert, and the same insert was producing caps with slightly undersized septa retention rings. The cosmetic defect was a leading indicator of a dimensional one. We reclassified septa retention as a major attribute and added a go/no-go gauge check to the outgoing inspection for that product line. The lesson: if a cosmetic defect is correlated with a tool wear pattern, it is not cosmetic. It is a process signal.

The decision rule I would write into every outgoing QC procedure

If you take one thing from this article, take this decision rule and put it in your outgoing inspection SOP: classify every attribute as critical, major, or minor before you pick a sample size, assign zero acceptance to critical, 0.65 AQL to major on sample-path attributes, 1.0 AQL to major on dimensional attributes, and 2.5 AQL to minor, use Z1.4 level II as the floor and level III for any attribute where a single failure causes a customer run failure, and record the sampling plan, the gauge ID, and the calibration date on every inspection record. That is the whole plan. Everything else is bookkeeping.

If you are sourcing chromatography products and want to see how a supplier’s outgoing QC holds up against these criteria, the product pages at Chromatography Products list the specifications and the test methods, and the manufacturing side at Aijiren publishes the batch record structure. For the underlying standards, ANSI/ASQ Z1.4 and ISO 2859-1 are the references to cite, and the EPA’s SW-846 test methods give you the analytical context for why a coring failure or a dimensional drift matters at the method level. Agilent’s GC consumables documentation, Thermo Fisher’s sample handling specifications, and Restek’s technical resources all publish the dimensional and performance specs your outgoing plan should be testing against. The sampling plan is the last chance to catch a defect before it becomes someone else’s failed run. Build it like it matters.

References & further reading

Related resources from our lab consumables network:

External references and standards: