Pick the inspection level from the consequence of the defect, not from habit

A 500-piece lot of 9 mm screw-thread caps failed outgoing QC at 2.4% major defects (cap height out of tolerance by 0.18 mm, enough to bottom out in a Shimadzu SIL-20A autosampler tray). The AQL plan called for 50 pieces at general inspection level II, AQL 1.0. Fifty pieces would have caught it about 70% of the time. Two hundred pieces caught it every time. That gap between what the plan promised and what the plan delivered is the whole argument for treating your hplc consumables AQL sampling plan as an engineering document, not a paperwork ritual.

Outgoing QC on consumables is not the same job as incoming inspection at the customer. By the time a vial, cap, septum, or ferrule reaches the loading dock, the process either made it right or it didn’t. Outgoing inspection is a gate, and a gate is only worth what its sampling plan is worth. Most suppliers run an AQL plan copied from a 1990s injection-molding manual and never revisit it. Here is how I build one that actually catches defects before they ship.

ANSI/ASQ Z1.4 (the U.S. adoption of ISO 2859-1) gives you three general inspection levels (I, II, III) and four special levels (S-1 through S-4). The default in almost every plan is Level II, and Level II is usually wrong for HPLC consumables in both directions.

Level II is designed for a producer who expects roughly 1% nonconforming product and wants a plan with reasonable consumer risk. For a cap that leaks at 0.5 mL/min under 100 bar, that consumer risk is a sequence lost, not a line stoppage. For a ferrule that seats fine but has a 0.05 mm burr, the consequence is a shoulder peak that shows up in run 40, not run 1. The defect severity, not the part count, should drive the level.

My rule set, refined over about a decade of shipping these parts:

• Critical defects (leak, chemical incompatibility, wrong polymer): Level III, AQL 0.065 or tighter. These are the ones that end a customer relationship.
• Major defects (dimensional out-of-tolerance, wrong thread pitch, wrong septum durometer): Level II, AQL 0.65 to 1.0.
• Minor defects (cosmetic flash, label misalignment, color variation): Level I, AQL 2.5 to 4.0.

The reason Level III matters for critical defects is the sample size. At a lot of 1,200 pieces, Level II gives you a code letter K sample of 125 pieces. Level III gives you code letter L, 200 pieces. That 75-piece jump moves your probability of detecting a 1% critical defect rate from about 71% to about 87%. If you are shipping to a pharma QC lab under a supplier quality agreement, that 16-point gap is the difference between passing and failing an audit finding.

AQL 0.065 is not a typo, and it is not a substitute for 100% inspection

New QC leads always ask why we use AQL 0.065 on critical defects instead of just inspecting every piece. The answer is that AQL 0.065 with a sample of 200 accepts zero defects. It is mathematically the same as 100% inspection of the sample, with the caveat that the sample is not the population. If you want zero defective caps in the box, you either 100% inspect (which fails on cost at 50,000 pieces per lot) or you build a process that is capable and use AQL 0.065 as the alarm.

That is the honest trade. AQL sampling is a statistical acceptance scheme, not a guarantee. A lot that passes at AQL 0.065 can still contain 0.3% defective parts. The reason we can ship at that rate is that the molding process capability (Cpk on cap height, typically 1.5 to 2.0 on a good tool) keeps the true defect rate well below the AQL. The AQL plan is the backstop, and the capability is the primary control. If a supplier quotes you AQL 0.065 and cannot show you a Cpk number, they are promising a gate they cannot back up.

Sample size is a table lookup, and the table is not the interesting part

Here is the working table for the lot sizes we actually run. This is the ANSI/ASQ Z1.4 single sampling plan, normal inspection, with the defect classes I use for vial and cap assemblies:

Two things the table does not tell you. First, the sample must be drawn from the lot at random across the production run, not from the top of the carton. If your molding cell runs four cavities and you sample the first 50 pieces off cavity 2, you have inspected one cavity. Draw from cartons pulled at the start, middle, and end of the shift. Second, the sample size is not the same as the inspection time. A 200-piece critical sample at 15 seconds per piece is 50 minutes of QC labor. Build that into your standard cost, or you will be tempted to shortcut it, and shortcutting it is how the 2.4% lot ships.

Defect classification is where most suppliers get lazy, and it is where the money is

The AQL number is only as good as the classification behind it. A defect class is a decision about consequence, and the decision has to be written down before the inspector sees the part, or every inspector will classify to their own tolerance.

Critical defects on HPLC consumables

• Leak under pressure: cap or vial assembly fails a 30-second pressure hold at 1.5x rated working pressure (typically 7 bar for a 2 mL vial cap, 100 bar for a ferrule fitting).
• Wrong polymer: a PTFE-lined cap that is actually silicone-lined, or a PEEK ferrule that is actually PTFE. Verify by DSC or FTIR on a pulled sample per shift, not per piece.
• Extractable contamination: a 2 mL water blank through the assembled vial shows a peak above 0.5 mAU at 210 nm that is not present in the control.
• Thread engagement failure: cap will not seat to the specified torque (typically 1.5 to 2.0 in-lb for a 9 mm screw cap, or 8 to 12 in-oz for a snap cap) without cross-threading.

Major defects

• Dimensional out-of-tolerance: cap height, vial neck OD, septum thickness outside the drawing. For a 9 mm cap, height tolerance is typically ±0.10 mm; a 0.18 mm excursion is a major defect even though it looks fine on the bench.
• Septum durometer out of range: a 70 Shore A silicone septum that measures 60 will puncture cleanly once and then coring on the second puncture. Measure with a durometer on 10 pieces per lot.
• Wrong thread pitch: a 9 mm cap on a 10-425 vial will thread on about 1.5 turns before it binds. The inspector has to test-fit, not just eyeball.

Minor defects

Cosmetic flash, label position, color variation between lots, minor surface scratches on the cap skirt. These do not affect function, but they affect customer perception, and a 4.0 AQL on minors keeps the plan honest without burning inspection time on a scratch.

The classification has to be written into the inspection instruction with a photo of each defect class. I have watched two inspectors disagree on whether a 0.05 mm burr on a ferrule was major or minor, and the disagreement cost a re-inspection of 4,000 pieces. The photo resolves it in five seconds.

What a realistic outgoing inspection record actually contains

Most suppliers show you a certificate of analysis with a single line: “Lot passed AQL inspection.” That is not a record. That is a claim. A record a pharma auditor will accept has, at minimum:

• Lot number and quantity, tied to the production date and the molding cell or assembly line.
• Sampling plan reference: standard (ANSI/ASQ Z1.4 or ISO 2859-1), inspection level, AQL per defect class, and the code letter.
• Sample size actually drawn and the draw method (start/middle/end of run, cartons sampled).
• Defect count by class (critical, major, minor) with the accept/reject numbers from the plan.
• Measured values for the dimensional checks: cap height, vial neck OD, septum thickness, with the gauge or caliper ID and calibration due date.
• Functional test results: leak test pass/fail, torque value at failure, puncture count to coring.
• Inspector ID, date, and disposition (accept, reject, rework, sort).
• Deviation record if any defect class exceeded its AQL, with the corrective action.

The puncture count is the one most suppliers leave out, and it is the one the customer’s method development group cares about most. A septum that survives 5 punctures at 22 gauge before coring is a different product from one that survives 2, even if both pass the dimensional AQL. If your outgoing record does not carry a puncture count, you are not recording the property that determines whether the customer’s autosampler runs a 96-well plate without a coring event.

Two field cases where the sampling plan was the fix

Case 1: the cap height drift that passed Level II

A customer running a Waters 2695 autosampler reported intermittent “vial not detected” errors, roughly one in 400 injections. The cap was a 9 mm screw cap with a 12 mm height. We pulled the customer’s retained lot and measured 200 caps: mean height 12.02 mm, standard deviation 0.06 mm, but a tail out to 12.19 mm. The autosampler’s gripper had a 12.15 mm clearance. About 2% of the lot was too tall.

The outgoing record showed AQL 1.0 at Level II, sample 50, zero majors. Fifty pieces from a normal distribution with a 2% tail will miss the tail about 36% of the time. The fix was not a tighter AQL. The fix was to add a 100% go/no-go height gauge at the assembly line exit (a 0.4-second check) and keep the AQL plan as the backstop for everything else. The customer’s error rate went to zero. The lesson: when a defect is a tail event on a continuous variable, AQL sampling is the wrong tool, and a 100% gauge is the right one. AQL is for attribute defects, not for a spec limit you can gauge in half a second.

Case 2: the septum that cored on the second puncture

A lot of 1,000 PTFE-faced silicone septa passed outgoing AQL at Level II, AQL 0.65, zero majors. Two weeks later a customer running an Agilent 7697A headspace sampler reported coring on the second puncture at 23 gauge. We pulled 20 retained septa and ran a puncture-to-coring test: 14 of 20 cored on puncture 2 or 3, against a spec of 5 minimum. The root cause was a durometer shift: the silicone measured 58 Shore A against a 68 to 72 spec. The AQL plan had no durometer check in it, so the defect class did not exist and the plan could not catch it.

The fix was to add durometer to the critical defect list and measure 10 pieces per lot, not per sample. Ten pieces at 5 seconds each is 50 seconds, and it catches a durometer shift that a dimensional AQL never will. We also added a puncture-to-coring test at 5 punctures on 5 pieces per lot. The outgoing record now carries a durometer range and a minimum puncture count, and the customer’s coring complaints stopped.

Torque, puncture count, and the numbers that belong on the spec sheet

If you are writing or auditing an outgoing QC plan, these are the parameters I would insist on seeing, with the values I use as defaults. Adjust for your specific part and customer method, but do not leave them unspecified.

• Cap torque: 1.5 to 2.0 in-lb for 9 mm screw caps; 8 to 12 in-oz for snap caps. Measure with a calibrated torque tester on 10 pieces per lot.
• Leak test: 30-second hold at 1.5x rated pressure, zero bubble or zero pressure decay. For a 2 mL vial cap, 7 bar; for a ferrule, 100 bar.
• Puncture to coring: minimum 5 punctures at 22 or 23 gauge before a visible core. Test 5 pieces per lot.
• Durometer: 68 to 72 Shore A for silicone septa; 55 to 65 for butyl. Test 10 pieces per lot.
• Dimensional: cap height ±0.10 mm, vial neck OD ±0.15 mm, septum thickness ±0.05 mm. Calibrated caliper, 20 pieces per lot.
• Extractables: 2 mL water blank at 210 nm, no peak above 0.5 mAU versus control. One composite per lot.

These are the numbers that belong in the outgoing inspection record, not just the pass/fail. A record that says “leak test passed” tells the auditor nothing. A record that says “leak test, 10 pieces, 7 bar, 30 s, zero failures” tells them the plan was executed.

Where the AQL plan stops and the process capability has to take over

I want to be blunt about the limit of sampling. An AQL plan with a sample of 200 and an accept number of zero will, on average, accept a lot with 1.5% defective parts about 5% of the time. That is the consumer risk, and it is baked into the standard. No sampling plan fixes a process that is drifting. The plan catches the drift; the process control prevents it.

So the outgoing AQL plan has to be paired with a capability number on the critical dimensions. If your cap height Cpk is 1.5, the tail beyond the spec limit is roughly 7 parts per million, and the AQL plan is a formality. If the Cpk is 0.9, the tail is 0.15%, and no AQL plan at Level II will protect the customer. I have seen suppliers quote AQL 0.65 and ship lots with a Cpk of 0.8 on the same dimension, and the two numbers contradict each other. Ask for both. If the supplier cannot produce the Cpk, the AQL plan is decoration.

The other limit is the link between outgoing inspection and the customer’s incoming inspection. If your customer runs their own AQL plan at Level II and you run yours at Level II, you are both sampling the same lot with the same blind spots. The customer’s incoming plan is not a substitute for your outgoing plan, and vice versa. The two plans should be complementary: you test the functional and extractable properties they cannot test on receipt, and they verify the dimensional and packaging properties you can only spot-check. That division of labor is worth writing into the supplier quality agreement. You can see how the receiving side handles this in the HPLC consumables reference material, and the supplier side of the same conversation lives at chromatographysupplier.com.

The outgoing inspection record is a sales document, not just a compliance document

Here is the part most QC managers miss. The outgoing inspection record is the first thing a procurement manager reads when they are deciding whether to switch suppliers. A record that shows the sampling plan, the sample size, the defect counts, the measured values, and the puncture count is a sales document. It says: we know what we are shipping, and we can prove it. A record that says “passed AQL” says nothing.

When I quote a new customer, I send the last three outgoing inspection records with the quote. Twice, that has closed a deal before the price conversation, because the customer’s own QC manager could see the plan was real. The OEM vial and vial closures lines both carry the same record format, and the format is the product as much as the part is.

The decision rule

Build the outgoing plan from the defect consequence, not from a default. Critical defects get Level III and AQL 0.065 with zero accept; major defects get Level II and AQL 0.65; minor defects get Level I and AQL 2.5. Draw the sample across the run, not from the top carton. Classify defects with a written instruction and a photo, and put the measured values, the durometer, the torque, and the puncture count on the record. Then pair the whole plan with a Cpk number on the critical dimensions, because a sampling plan that is not backed by process capability is a gate that opens on average. If the supplier cannot show you both the plan and the Cpk, the plan is paperwork. The standards to cite are ANSI/ASQ Z1.4 and ISO 2859-1, and the reference methods for the functional tests are in U.S. EPA SW-846 for container performance and in the Restek technical resources for septum and liner behavior. For instrument-side fit and compatibility, the Agilent consumables documentation and the Thermo Fisher chromatography sample handling pages give the dimensional clearances you need to set the go/no-go gauges. The plan is only as good as the gauge behind it.

References & further reading

Related resources from our lab consumables network:

• Hplc Consumables at hplcconsumable.com
• Chromatography Supplier at chromatographysupplier.com
• Oem Vial at oemsamplevials.com
• Vial Closures at vialclosures.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