AQL 2.5 on a critical defect is a pass you will regret in six months--Reagent Bottle for Sale

A 12,000-bottle lot of 250 mL amber soda-lime reagent bottles came off the pallet with 34 units that failed a 0.3 mL water fill test at the shoulder seam, and the supplier’s outgoing record said “AQL 2.5, pass.” It passed because the plan sampled 80 pieces against a critical defect class that should have been AQL 0.065, and the inspector had been classifying seam leaks as major. That is the whole problem with reagent bottle AQL sampling plans and outgoing QC in one shipment: the number on the certificate is only as good as the defect class it was applied to. If the class is wrong, the AQL is decoration.

What follows is how I build and audit an outgoing sampling plan for reagent bottles, what sample sizes actually mean at the volumes this industry ships, how to classify defects so the plan has teeth, and what a real inspection record should contain when a procurement manager asks for it. It assumes you already know the product. It does not re-derive incoming glass inspection or in-process capability, because those live elsewhere on this site.

ANSI/ASQ Z1.4 (the U.S. adoption of ISO 2859-1) lets you pick an AQL for each defect class independently. Most reagent bottle suppliers default every class to 2.5 because it is the middle of the table and the inspector does not have to think. That is the failure mode. The AQL should be set by what the defect does downstream, not by what the supplier finds convenient.

My working classification for reagent bottles, refined over about a decade of customer complaints:

Critical (AQL 0.065 or 0.10): Anything that causes leakage, chemical exposure, or cross-contamination. Seam pinholes that leak under 0.3 mL water fill. Thread defects that prevent torque holding. Missing or wrong liner. Cap that will not seal at the specified torque. Glass chips inside the cavity. This class gets the tightest AQL because a single escape is a customer incident, not a cosmetic complaint.
Major (AQL 1.0 or 1.5): Defects that make the bottle unusable in the intended application but do not create a safety event. Wrong neck finish (GPI 24-400 shipped as 24-410). Graduation error beyond ±2% of nominal at the stated fill point. Visible cords or stones in the wall. Chipped bead that will not seal. Amber color outside the transmission spec at 400 nm.
Minor (AQL 2.5 or 4.0): Cosmetic or handling issues that do not affect function. Scuffs on the shoulder. Light label-print misregistration. Slight mold-parting line flash that does not cut a glove. These can run at 4.0 without much risk.

The trap is reclassifying a critical defect as major to keep the lot moving. I have seen seam leaks downgraded because “the customer will catch it at fill.” That logic assumes the customer has a 100% leak test, which almost none do. If your outgoing plan classifies a leak as major at AQL 1.5, you are accepting up to 1.5% leaking bottles at the accept/reject boundary. On a 12,000-piece lot that is 180 leaking bottles shipped with a “pass” stamp.

Sample size follows lot size and inspection level, not the supplier’s habit

The single most common question I get from procurement is “how many do you inspect?” The honest answer is “it depends on the lot size and the inspection level, and if the supplier cannot tell you which general inspection level they use, they are not running a plan.”

Under Z1.4, General Inspection Level II is the default for most consumer and industrial goods. Level I is roughly half the sample size and is defensible for mature, stable processes with a good history. Level III doubles it and is what I use for a new supplier, a new mold, or the first three lots after a tooling change. Special levels S-1 through S-4 are for destructive or very expensive tests and are not appropriate for visual outgoing inspection of bottles.

Here is what the numbers look like for reagent bottle lot sizes, General Level II, single sampling, normal inspection:

Lot size	Sample size (code letter)	Accept at AQL 1.0	Accept at AQL 2.5
501 to 1,200	80 (J)	2	5
1,201 to 3,200	125 (K)	3	7
3,201 to 10,000	200 (L)	5	10
10,001 to 35,000	315 (M)	7	14
35,001 to 150,000	500 (N)	10	21

Two things to notice. First, at AQL 0.065 the acceptance number is zero at every sample size in the table. That means any critical defect in the sample rejects the lot. This is why the critical class needs a real plan, not a token one: one leaking bottle in 315 is enough to hold the lot, and that is the correct outcome. Second, at AQL 2.5 on a 10,000-piece lot you can accept 10 minor defects in 200 pieces. That is fine for scuffs. It is not fine for anything a chemist will notice at the bench.

If the supplier quotes “AQL 2.5” without naming the inspection level, the sample size, and the defect class, the plan is incomplete. Ask for the code letter. A supplier who cannot produce it is running a visual walk-around, not a sampling plan.

Defect classification is where the plan actually lives or dies

I keep a written defect classification sheet for every reagent bottle SKU, and the inspector works from it, not from memory. The sheet lists each defect, the class, the measurement method, and the accept/reject threshold. Without that sheet, two inspectors on the same line will classify the same bottle differently, and the AQL math becomes noise.

Specific classifications I use, with the measurement that decides them:

Leak and seal defects (critical)

A 0.3 mL water fill, cap torqued to the SKU’s spec (typically 0.9 to 1.4 N·m for a 24-400 PP cap with a PTFE-faced liner, 1.4 to 2.0 N·m for a 38-400), inverted on absorbent paper for 10 minutes at ambient. Any wicking is a reject. This test is cheap and it catches the seam and thread defects that visual inspection misses. I have seen a lot pass visual at 315 pieces and fail 8 of 50 on water fill because the mold close on the shoulder seam was running 0.04 mm thin.

Neck finish and dimensional defects (major)

GPI 24-410 thread pitch and T-dimension checked with a go/no-go gauge. T-dimension tolerance for a 24-410 is typically ±0.38 mm; for 38-400 it is ±0.51 mm. Out-of-gauge is major, not minor, because a cap that does not seat will leak under the customer’s fill line even if it passed your water fill test at a different torque.

Capacity and graduation (major)

Fill to the stated fill point with deionized water at 20 °C and weigh. Tolerance is ±2% of nominal for Class B ware; tighter for Class A volumetric. The “capacity” line on the carton is nominal until the fill point is stated, and a bottle that holds 248 mL at the 250 mL line is a major defect if the customer’s method assumes 250.

Chemical resistance and color (major)

Amber transmission at 400 nm checked against the SKU spec (typically ≤10% transmission at 400 nm for amber 320 glass). Clear bottles checked for visible tint. This matters for photodegradable analytes and for anyone running the bottle through an autosampler workflow.

Cosmetic (minor)

Scuffs, parting-line flash under 0.2 mm, print misregistration under 1 mm. None of these affect function.

What a realistic outgoing inspection record actually contains

Most “certificates of analysis” I see for reagent bottles are one page of boilerplate with a pass stamp. A record that will survive an audit and actually tell you something has these fields, per lot:

Lot/batch number and date of manufacture, traceable to the mold cavity if the supplier can do it (good ones can).
SKU, nominal capacity, neck finish, and material (soda-lime, borosilicate 3.3, PP, etc.).
Sampling plan reference: standard (Z1.4), inspection level (e.g., General II), sample size and code letter (e.g., 200, L), and the AQL per defect class.
Defect counts by class, not just pass/fail. “Critical 0/200, major 2/200, minor 6/200” tells you the lot is drifting even when it passes.
Test methods and instruments: water fill volume and torque value, go/no-go gauge ID, spectrophotometer for amber transmission, balance for capacity check.
Measured values for the quantitative checks, at least min/max/mean for capacity and torque, with the RSD. A torque RSD above 15% on a capping line is a warning sign even if every unit is in spec.
Inspector ID and date, plus the disposition (accept, accept with deviation, reject).
Deviation log for any lot accepted with a documented concession.

If the supplier’s record has none of the defect counts and no sample size, ask for the underlying inspection sheet. The certificate is a summary; the sheet is the evidence. A supplier who cannot produce the sheet within a day is not running the plan they are certifying.

Two field cases where the sampling plan caught what visual missed

Case 1: the amber transmission drift

A customer running photodegradable standards started seeing degraded peaks after four weeks in storage. The bottles were amber 250 mL, and the supplier’s outgoing record showed “color: pass, visual.” We pulled 30 bottles from three lots and measured transmission at 400 nm on a UV-Vis. Two of the three lots were at 14 to 17% transmission against a spec of ≤10%. The supplier had been classifying color as a minor visual defect and sampling 80 pieces at AQL 4.0, which let the drift through. The fix was to move amber transmission to the major class, add a spectrophotometer check at 400 nm to the outgoing plan at AQL 1.0 with sample size 125, and require the measured transmission range on the certificate. The next three lots came in at 6 to 9% and the customer’s peak degradation stopped.

Case 2: the torque that passed at the supplier and failed at the filler

A lot of 38-400 borosilicate bottles passed the supplier’s water fill at 1.6 N·m. At the customer’s fill line, running at 1.9 N·m on an automatic capper, 4% of bottles leaked at the thread root. The supplier’s plan had tested torque at a single value, not across the range. We re-ran the outgoing check at three torque values (1.4, 1.7, 2.0 N·m) on 50 pieces per lot and found that the T-dimension was running at the low end of tolerance, so the cap bottomed out before the liner compressed at the higher torque. The fix was a go/no-go gauge check at the T-dimension lower limit added to the major class, plus a note on the certificate stating the tested torque range. The leak rate went to zero on the next lot.

Switching from normal to tightened inspection is a decision, not a default

Z1.4 includes switching rules: after two out of five consecutive lots are rejected on original inspection, you move to tightened inspection, which uses a larger sample size for the same lot and a lower acceptance number. Most suppliers never invoke it because they never track the consecutive-lot history. If your supplier cannot tell you their current inspection status (normal, tightened, or reduced), the switching rules are not being applied, and the AQL on the certificate is a snapshot with no memory.

I require the inspection status on the certificate. It is a one-word field and it tells you whether the supplier is paying attention to their own history. A supplier stuck on tightened for six months is telling you something about the process. A supplier on reduced inspection after a stable run is telling you something good, but only if the reduced plan still samples enough to catch a critical defect, which at AQL 0.065 it does (acceptance number stays at zero).

Reduced inspection is fine for minor defects and dangerous for critical ones

Reduced inspection cuts the sample size to roughly 40% of normal. For cosmetic and minor defects on a mature process, that is a reasonable trade. For critical defects, the acceptance number at AQL 0.065 is already zero, so reducing the sample size just lowers your probability of catching the one leaking bottle in the lot. I do not allow reduced inspection on the critical class, ever. The supplier can reduce the minor class after 10 consecutive accepted lots; the critical class stays at normal or tightened.

This is also where the choice of inspection level interacts with the defect class. A supplier on General Level I with AQL 2.5 across the board is sampling 32 pieces on a 3,000-piece lot. That is not enough to catch a 1% leak rate with any confidence. The math is unforgiving: to detect a 1% defect rate with 95% confidence you need roughly 300 pieces with zero accepts. If the supplier’s plan samples 32, they are detecting defects at maybe 9% prevalence. That is a plan for a mature, stable process, not for a new mold.

The bottle, the cap, and the liner are one inspection unit

One thing I insist on that many suppliers miss: the outgoing inspection must test the bottle with the cap and liner it ships with, at the torque the customer will use. A bottle that passes water fill with a supplier’s in-house cap and fails with the customer’s cap is a bottle that was never really tested. The liner material matters too: a PTFE-faced silicone liner and a plain PE liner have different compression set, and the torque that seals one will not seal the other. If you are buying bottles and caps from different suppliers, the outgoing plan at the bottle supplier cannot validate the seal. That validation belongs at your incoming inspection, which is a different article, but the point is that the outgoing record should state which cap and liner were used in the water fill test.

For anyone running these bottles into an autosampler workflow, the same logic applies to the downstream consumables. The HPLC autosampler vial and the sample vial have their own AQL plans, and a bottle that passes at AQL 0.065 can still fail at the vial transfer step if the neck finish is off. The reagent bottle is one link in a chain that includes the chromatography products around it.

Troubleshooting checklist for a plan that is not catching defects

If lots are passing outgoing inspection and failing at the customer, work through this in order:

Check the defect class first. Is the leak classified as critical or major? If major at AQL 1.5, that is your answer.
Check the sample size against the lot size. A 200-piece sample on a 50,000-piece lot is code letter L, which is correct for General II up to 10,000. Above that you need code M (315) or N (500).
Check the inspection status. If the supplier is on normal after two rejections in five lots, the switching rules are not being applied.
Check the test method. Water fill at one torque value, or visual only, will miss thread and seam defects. Add the three-torque check and the go/no-go gauge.
Check the cap and liner used in the test. If it is not the shipping configuration, the seal test is not valid.
Check the record. If there are no defect counts by class and no sample size, there is no evidence the plan ran.
Check the quantitative fields. Capacity mean and torque RSD on the certificate will show drift before the lot fails.

For reference material on the test methods themselves, the U.S. EPA SW-846 test methods cover container and sampling requirements that inform how strict the plan needs to be for environmental work, and the major instrument vendors publish sample-handling guidance that is worth reading alongside your own plan: Agilent’s GC columns and accessories, Thermo Fisher’s chromatography and sample handling, and Restek’s technical resources. None of them will write your AQL plan for you, but they set the context for why the container matters.

The decision rule I give every procurement manager

Set the critical defect class at AQL 0.065 with a zero acceptance number and a sample size of at least 200 (code letter L) on any lot over 3,200 pieces, and do not allow reduced inspection on that class. Classify leaks, thread defects, wrong liners, and internal chips as critical, and put the water fill at the customer’s torque plus the go/no-go gauge on the outgoing sheet. Require the certificate to carry the sample size, the code letter, the defect counts by class, the inspection status, and the measured capacity and torque with RSD. If the supplier cannot produce those fields, the “AQL 2.5, pass” stamp is a number without a plan behind it, and the 34 leaking bottles are already on the pallet.

References & further reading

Related resources from our lab consumables network:

External references and standards: