Last March a lot of 40,000 9 mm screw-top vials shipped against an outgoing record that read “AQL 1.0, pass.” The receiving lab found 11 vials in the first 500 with thread flash that would not seal on a HPLC vial position at 1,200 psi. Nobody had defined what “pass” meant, nobody had classified the flash as a defect, and the sampling plan had been copied from a carton supplier’s certificate. That is the failure mode this article is about: an analytical testing AQL sampling plan that exists on paper but does not describe what you are actually inspecting, how many you inspect, or what counts as a reject.

Outgoing QC for analytical consumables is not the same problem as outgoing QC for a molded toy. The customer is going to put your part into a 40 °C autosampler tray, pierce it with a 0.63 mm needle, and read a chromatogram where a 0.5% area RSD shift is a finding. The sampling plan has to be built around that reality, not around a generic ISO 2859 table pulled off the internet.

AQL is a property of a defect class, not of a product. If you write “AQL 1.5” on a certificate without saying which characteristic it applies to, the number is decoration. The first decision in any outgoing plan is the classification.

Critical, major, minor, and the fourth class most plans forget

I use four classes, not three. The fourth is what I call a functional-metrological defect: the part passes a visual check but will not perform analytically. It is the class that gets missed because inspectors are trained to look, not to measure.

• Critical (AQL 0, zero accept): anything that creates a safety or contamination hazard. Glass shards in a bag, silicone lubricant on the inside of a vial, a cap with a pierced septum shipped as new, particulate visible at 10x in a clean vial. You inspect 100% or you accept the risk in writing. For critical defects I use c=0 with a sample size tied to the lot, typically 125 pieces for lots over 10,000.
• Major (AQL 0.65 to 1.0): a defect that stops the part working. Thread flash that prevents a seal, a septum with a durometer reading outside 45 ± 5 Shore A, a cap that will not hold torque at 1.5 in-lb, dimensional stack failure on the 9-425 neck (GPI 9 mm finish, 8.4 mm ID, 15.0 mm OD at the thread crown).
• Minor (AQL 2.5 to 4.0): cosmetic or handling issues that do not stop function. Scuffs on the outside of a vial, a label printed 0.5 mm off-center, a bag seal that is neat but not pretty.
• Functional-metrological (AQL 0.65): the characteristics that decide whether the part performs in the instrument. Septum puncture force, reseal after 3 punctures at 23 °C, extractables screening by EPA SW-846 method 8270 on a representative composite, background at 210 nm on a UV-transparent vial. These are not visual, so they need a different sampling cadence.

The classification drives everything downstream. A plan that treats a septum durometer excursion as a minor defect will ship bad lots with a clean certificate. I have seen exactly that on a 20 mm crimp cap where the butyl disc ran 52 Shore A and the crimper cracked the aluminum at 1.4 in-lb instead of the specified 1.1 in-lb.

Pick the sample size from the lot and the class, then stop arguing about it

ISO 2859-1 general inspection level II is the default most buyers expect, and for good reason: it is defensible, it is on the certificate, and it maps to a known consumer’s risk. For a lot of 10,001 to 35,000 pieces, level II gives you a code letter M and a sample size of 315. For 3,201 to 10,000, code letter L, sample size 200. For 501 to 1,200, code letter H, sample size 80.

Those numbers are not arbitrary. They come from the operating characteristic curve: at AQL 1.0 and n=315, the probability of accepting a lot at exactly 1.0% defective is about 95%, and the probability of accepting a lot at 5% defective is under 5%. That is the trade you are making. If you want tighter protection, you either raise the sample size or you tighten the AQL. You cannot get both for free.

Two things I insist on beyond the table. First, for functional-metrological characteristics, the sample size is capped by test cost, not by the table. You are not going to run 315 septum puncture tests on a lot. You run 5 or 10, and you treat them as a separate attribute plan with c=0. Second, for critical defects, you never use the table’s Ac number. Ac=0, always, and the sample size is a floor, not a ceiling.

When to switch to tightened or reduced inspection

ISO 2859-1 has a switching rule most suppliers ignore. Two out of five consecutive lots rejected on original inspection, and you go to tightened inspection (one code letter up, same AQL). Five consecutive lots accepted on tightened, and you return to normal. Ten consecutive lots accepted on normal, and you may go to reduced (one code letter down, sample size roughly 40% of normal, Ac same). I have seen a supplier run reduced inspection for 18 months because nobody updated the plan after the first five passes. Reduced inspection on a lot of 40,000 turns a sample of 315 into a sample of 125. That is a decision, and it needs to be documented, not drifted into.

What you actually measure, and with what instrument

A visual inspector with a 10x loupe can find thread flash, scuffs, and missing septa. It cannot find a 3 Shore A durometer shift, a 0.08 mm dimensional drift on the neck ID, or a background absorbance of 0.04 AU at 210 nm. Those need instruments, and the outgoing plan has to name them.

Dimensional and mechanical checks

• Neck ID and thread pitch: go/no-go gauges on the GPI 9 mm finish. Pin gauge at 8.40 mm minus 0.05 mm, ring gauge at 15.00 mm plus 0.05 mm. Check 20 pieces per sample, not 5. Dimensional drift on a molding cavity shows up in the tail of the distribution before it shows up in the mean.
• Septum durometer: Shore A per ASTM D2240, 5 readings per septum, 5 septa per sample. Spec 45 ± 5 for standard silicone, 55 ± 5 for PTFE-faced. Record the mean and the range.
• Puncture force and reseal: 0.63 mm needle, 3 punctures at 23 °C and 50% RH, force recorded on a 50 N load cell. Spec typically 2.5 to 4.5 N per puncture on a 9 mm silicone septum. Reseal checked by pressure decay at 1.0 psi over 30 s on a 2 mL fill.
• Cap torque: applied torque on the capper, measured back-off torque on 10 pieces. Spec 1.3 to 1.6 in-lb for a 9-425 screw cap with a PTFE-faced silicone liner. Below 1.1 in-lb, you get evaporation over a 24-hour tray hold. Above 1.8 in-lb, you crack the cap skirt and the seal fails anyway.

Analytical background checks

This is where most outgoing plans are thin. A representative composite of 20 vials from the lot, extracted with 1 mL of HPLC-grade acetonitrile for 30 minutes at 40 °C, injected on a C18 column at 254 nm. Acceptance: no peak above 0.005 AU at the retention time of common plasticizers (DEHP at ~8.4 min on a standard 150 × 4.6 mm, 5 µm column), and total integrated area under 0.02 AU·min. For UV-transparent vials, the cut is 0.04 AU at 210 nm against a water blank. The reference method is your own, but the acceptance criteria should be traceable to what your customer’s method can tolerate. If you do not know their detection limit, ask.

For GC-facing products, a background run on a 30 m × 0.25 mm × 0.25 µm column with a 1 µL splitless injection of the lot composite is the check. Any siloxane peak above 5% of the internal standard area is a reject. The Restek technical library has good reference chromatograms for what a clean septum background should look like, and I use them as the visual benchmark in inspector training.

Two field cases where the plan caught what the eye missed

Case 1: the 9-425 lot that passed visual and failed at 40 °C

A customer running a stability study at 40 °C / 75% RH reported area RSD of 4.2% on triplicate injections from the same vial over a 48-hour tray hold. The visual outgoing record was clean. We pulled the retained sample and ran the dimensional check: neck ID was 8.31 mm on 6 of 20 pieces against a spec of 8.40 mm minus 0.05. The cap seated, the torque measured in spec, but the reduced interference meant the liner did not compress evenly, and the vial lost 0.8 mg of solvent over 48 hours. The measurement that identified it was the pin gauge, not the leak test. The fix was a cavity-specific insert replacement on the two worst cavities and a tightening of the outgoing dimensional sample from 5 to 20 pieces per lot. The AQL for that characteristic went from 1.0 to 0.65, and the sample size stayed at code letter M because the cost of the extra 15 gauge checks was trivial next to a repeat stability study.

Case 2: the septum durometer drift that read as a method problem

A lot of 22 mm crimp caps produced a ghost peak at 6.2 min on a customer’s EPA 8260 method. The customer’s first call was to the instrument, then to the column. The outgoing record showed durometer “in spec” because the inspector had measured 3 pieces, all at 47 Shore A. We re-measured 20 pieces and found a bimodal distribution: 15 pieces at 46 to 48, and 5 pieces at 53 to 55. The high-durometer discs were from a second cavity that had run with a short cure. At 53 Shore A, the crimper at 1.1 in-lb did not fully deform the aluminum skirt, and the septum leaked siloxane into the inlet. The fix was a cure-time interlock on that cavity and a rule that durometer samples must span all cavities, 2 pieces per cavity, minimum 10 pieces. The ghost peak disappeared on the retest of the retained lot. If you want a deeper treatment of that failure mode, our chromatography consumables reference covers the material side.

What a realistic outgoing inspection record contains

The certificate that ships with a lot should be reproducible by a third party from the record alone. Most are not. Here is the minimum I will accept on a supplier’s outgoing record, and what I generate on my own lines.

• Lot identification: lot number, cavity range or press assignment, resin lot, molding date, and the operator and inspector initials. If you cannot trace back to a cavity, you cannot do a targeted recall.
• Plan reference: the standard (ISO 2859-1), the inspection level (II), the AQL per defect class, and the sample size actually used. “AQL 1.0” alone is not a plan reference.
• Characteristic list with the instrument and the calibration date. A durometer reading without the gauge serial number and its last calibration date is an opinion.
• Raw readings, not just pass/fail. The individual dimensional values, the durometer mean and range, the puncture force per piece, the background absorbance. If the record shows only “pass,” you cannot detect a drift on the next lot.
• Defect tally by class. Number inspected, number of critical, major, minor, and functional-metrological defects, and the accept/reject decision per class.
• Disposition of rejects. Reworked, scrapped, or downgraded to a non-analytical grade. If downgraded, the record must say so, and the downgraded lot must not ship on the analytical certificate.
• Retained sample location and quantity. Minimum 20 pieces per lot, held for the shelf life plus 6 months. Retained samples are what let you answer the customer’s 48-hour complaint without a full re-run.

I also want the record to name the person who made the accept decision, not just the inspector who took the readings. Those are different roles and should be different signatures. On a lot where a functional-metrological characteristic is close to the limit, the accept decision is a judgment call and it needs a name attached.

Sampling from a bag is not sampling from a lot

One operational point that gets skipped: if your 315-piece sample all comes from the top of a single carton, you have sampled a carton, not a lot. Stratify. Pull from the first, middle, and last carton of the lot, and from at least three positions within each carton. For a lot of 40,000 in 40 cartons of 1,000, that means 8 cartons minimum and roughly 40 pieces per carton. The inspector who grabs 315 from the nearest pallet is measuring the pallet.

Same rule for functional-metrological samples: one piece per cavity, minimum 10 pieces, drawn across the production window. If the lot was molded over 6 hours, take samples from hour 1, hour 3, and hour 6. Cure drift and mold-temperature drift are time-dependent, and a sample drawn at hour 5 only tells you about hour 5.

Where the plan meets the customer’s method

The outgoing plan is not a standalone document. It has to be calibrated to what the customer’s method can tolerate. A vial that passes your 0.04 AU background check at 210 nm will still fail a customer running at 195 nm on a short-pathlength detector. A septum that reseals after 3 punctures will fail a customer running a 96-position tray with 4 injections per vial. The Agilent consumables documentation and Thermo Fisher’s sample handling references both publish the operating envelopes their instruments expect, and those envelopes are a good starting point for setting your acceptance limits. If your customer’s method is tighter than the envelope, your plan has to be tighter too, and that is a conversation to have before the first shipment, not after the first complaint.

For a broader view of how the outgoing plan sits inside the full testing workflow, see our analytical testing reference, and for the vial-side dimensional detail, the lab vials library has the GPI and ISO 4796 drawings that the gauges are built to.

Troubleshooting checklist: when the plan is not catching what it should

• Complaints cluster on one characteristic. Check whether that characteristic is in the plan at all, and whether it is classified as minor when it should be functional-metrological.
• Complaints cluster on one cavity or one press. The sample is not stratified. Add cavity coverage and time-window coverage.
• Pass rate is 100% lot after lot. Either the process is genuinely capable or the sample is too small. Run a capability check (Cpk) on the dimensional characteristic; if Cpk is above 1.33, the sample is probably fine. If Cpk is below 1.0 and the plan still passes every lot, the plan is not measuring the characteristic.
• Reject rate jumps after a resin change. The resin lot is a variable in the plan. Add resin lot to the record and require a first-article check on any resin change.
• Customer finds defects your inspector missed. Compare the defect definitions line by line. In my experience, 80% of “missed” defects are defects your plan never defined.

The decision rule is simple: classify every characteristic that matters to the customer’s method as functional-metrological, set its AQL at 0.65 or tighter with c=0 on the analytical checks, stratify the sample across cavities and time, and write the raw readings into the record so the next lot can be compared against this one. If a characteristic cannot be measured with a calibrated instrument and written down as a number, it is not in the plan, and a lot that ships without it is shipping on hope.

References & further reading

Related resources from our lab consumables network:

• Analytical Testing at testinganalytical.com
• Hplc Vial at vialforhplc.com
• Chromatography Consumables at chromatographyconsumables.com
• Lab Vials at labvials.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