AQL 0.65 on critical defects is the right floor for a 2 mL autosampler vial

A 40,000-vial lot of 2 mL clear screw-thread vials failed outgoing inspection last March on a single critical defect: one vial in the AQL sample had a thread root crack that leaked at 0.6 bar in the pressure-decay test. Under the plan in force at the time (AQL 0.65 critical, sample size code letter M, n=315), one critical defect is a reject. The lot went back to the molding floor, and the rework cost more than the entire margin on the order. That is what a lab vials AQL sampling plan and outgoing QC actually decides, every lot, whether the shipment leaves or the pallet turns around.

Most of what gets written about vial quality is about the vial itself. This is about the paperwork and the sampling arithmetic that stands between the molding floor and the customer’s loading dock. If you are setting up or auditing an outgoing inspection scheme for vials, caps, and septa, the pieces that matter are the AQL level, the sample size, the defect classification, and the record that survives the audit. Get those four wrong and you will either ship bad lots or quarantine good ones.

The AQL level is a decision about how many defective units you are willing to accept in a lot before you call it bad. It is not a quality target. It is the upper bound on the process average you are willing to live with at the outgoing gate.

For a 2 mL autosampler vial, I run three defect classes with three AQLs:

  • Critical (AQL 0.65): defects that cause a leak, a lost sample, or an instrument fault. Thread root cracks, incomplete thread formation, septum punctured through to the liner, wrong septa material on a certified lot.
  • Major (AQL 1.5): defects that degrade performance without immediate failure. Dimensional out-of-tolerance on crimp neck OD, ovality beyond spec, visible glass inclusions, closure torque below the minimum.
  • Minor (AQL 4.0): cosmetic and handling defects. Scratches, label misalignment, minor color variation in amber glass, cap knurling imperfections.

Why 0.65 and not 1.0 on critical? Because at 1.0 the acceptance number for sample size code letter M (n=315) is 7. Seven leaking vials in a sample of 315 passes the lot. At 0.65 the acceptance number is 5. That is the difference between shipping a lot with a leak rate near 2% and one near 1%. For a lab running 500 injections a day, the difference is a needle-in-the-wrong-place event every few weeks versus every few months. The 0.65 level is the same one used across much of the medical device and pharmaceutical packaging supply chain, so it survives a customer audit without a fight.

For crimp vials going into a regulated method, I have pushed some customers to AQL 0.25 on critical. That is defensible when the downstream method is USP <661> compliant and the customer is running a stability study. It is not necessary for general-purpose autosampler work. Match the AQL to the consequence, not to the customer’s mood.

Sample size code letter M is the workhorse, and the table is not optional

Sample size follows ISO 2859-1 (the same scheme as ANSI/ASQ Z1.4). You pick a code letter from lot size and inspection level, then read the acceptance and rejection numbers from the AQL table. Do not eyeball it. The numbers are not linear and the acceptance numbers do not scale the way people assume.

For lot sizes between 10,001 and 35,000 units at General Inspection Level II, the code letter is M, sample size 315. For 35,001 to 150,000, code letter N, sample size 500. For 3,201 to 10,000, code letter L, sample size 200. For 501 to 1,200, code letter H, sample size 80.

Lot size Code letter Sample size (n) Ac/Re at AQL 0.65 Ac/Re at AQL 1.5 Ac/Re at AQL 4.0
501 to 1,200 H 80 0 / 1 1 / 2 7 / 8
1,201 to 3,200 J 125 0 / 1 3 / 4 10 / 11
3,201 to 10,000 L 200 1 / 2 5 / 6 14 / 15
10,001 to 35,000 M 315 3 / 4 7 / 8 21 / 22
35,001 to 150,000 N 500 5 / 6 10 / 11 21 / 22

Read the Ac/Re column carefully. At AQL 0.65 on code letter H (n=80), the acceptance number is zero. Any critical defect rejects the lot. At code letter M (n=315), you can accept up to three. That is counterintuitive to people who expect larger samples to be stricter. They are stricter in the statistical sense (a larger n detects a smaller true defect rate), but the acceptance number goes up because the sample is larger. The lot is judged on the same process average.

One practical note: when the customer’s purchase order specifies a sample size but not a code letter, ask which inspection level it came from. I have seen POs specify “n=125” without saying whether that was Level II at 1,201 to 3,200 units or Level I at 10,001 to 35,000. The acceptance numbers differ. Get it in writing before the first shipment.

Critical, major, and minor are defined by the test, not by the inspector’s mood

The single biggest failure I see in outgoing inspection records is a defect classification that depends on who is on shift. If the inspector decides whether a scratch is minor or major based on how the day is going, the AQL plan is theater.

Every defect class has to be tied to a physical test with a pass/fail threshold. For vials, the critical list is short and mechanical:

  • Leak at pressure: pressure-decay test at 0.6 bar for 30 seconds, leak rate above 1×10⁻⁴ mbar·L/s is a fail. This catches thread root cracks, incomplete seal at the cap-to-neck interface, and septum defects that only show under pressure.
  • Septum puncture integrity: the septum must hold a seal after three punctures with a 22-gauge needle at the specified injection depth. A fourth-puncture leak is a major, not a critical, because most methods do not exceed three injections per vial.
  • Closure torque: applied torque at the capping station must be within 0.9 to 1.2 N·m for a 9 mm screw cap. Below 0.9 N·m the closure can back off in the autosampler tray. Above 1.2 N·m the septum can deform and the next puncture can coring. Both directions are critical.
  • Dimensional stack: crimp neck OD within ±0.15 mm of nominal, body OD within ±0.20 mm, height within ±0.30 mm. Out-of-tolerance on any of these is major, not critical, unless it prevents the vial from seating in the autosampler.

Minor defects do not need a test method, but they need a written definition with a photo standard. Scratches longer than 3 mm, chips at the base, visible bubbles in the glass wall. The photo standard is what makes the classification reproducible across inspectors and across shifts. Without it, you are back to mood-based inspection.

What a realistic outgoing inspection record actually contains

A record that survives an audit has a fixed field set. Not a paragraph of narrative. A structured form with the same fields on every lot, every time. Here is what I put on ours, and what I look for when I audit a supplier:

  • Lot identification: internal lot number, molding date, molding machine ID, resin lot number for the glass tubing, septum lot number, cap lot number.
  • Sampling plan reference: the standard (ISO 2859-1), inspection level, code letter, sample size, and the AQL levels applied to each defect class.
  • Sample pull record: which pallets and which cartons the sample came from, with the sample drawn across the lot rather than from the top carton. I pull from at least five cartons per lot, stratified by pallet position.
  • Test results: the measured values, not just pass/fail. Pressure-decay leak rate per unit, torque per unit, dimensional readings per unit. A record that says “pass” with no numbers is not a record.
  • Defect log: every defect found, with the defect code, the classification, and the count. Even accepted lots need this, because the trend across lots is what tells you the process is drifting before it fails.
  • Disposition: accept, reject, or accept-with-deviation (with the deviation documented and signed). Rejected lots get a rework record and a re-inspection record.
  • Inspector and date: name, not initials, and the date the inspection was completed. Not the date the form was printed.

The record I reject most often from suppliers is one where the defect log is empty on every lot. That is not a sign of a perfect process. That is a sign that nobody is actually looking. A real process has a nonzero minor defect count on most lots, and a major defect every few lots. If the log is always clean, the inspection is not happening.

For the septa specifically, the record has to carry the material certification. A PTFE-faced silicone septum is not the same as a pure silicone septum, and the outgoing record should reference the septa lot and the material certificate. If the customer is running a method that depends on the PTFE face, a substitution at the capping station is a critical defect even if the septum passes every physical test.

Two field cases where the sampling plan caught what the process missed

Case 1: the 0.65 AQL caught a thread root crack that the in-process gauge missed

Symptom: a customer running a VOA vial method reported intermittent leaks in the autosampler tray, roughly one in 400 vials, always on the same lot. The in-process gauge at the molding station showed no dimensional drift. The outgoing record for that lot showed a clean defect log.

What identified it: we pulled the retained sample from the outgoing inspection and ran the pressure-decay test at 0.6 bar. Six of the 315 retained vials leaked. The leak rate was just above the 1×10⁻⁴ threshold, which is why the in-process gauge (which only checked dimensions, not pressure) never saw it. The crack was at the thread root, below the surface, and only opened under pressure.

The fix: added pressure-decay to the outgoing critical test for every lot, not just a quarterly audit. The molding floor added a dye-penetrant check at the thread station. The lot was reworked and re-inspected, and the customer’s leak rate dropped to zero on subsequent lots. The lesson: dimensional inspection does not catch a crack. Pressure does.

Case 2: the torque spec was right, the capping station was wrong

Symptom: a lot of 9 mm screw caps passed outgoing inspection on dimensions and leak, but the customer reported septum coring on the first injection in about 2% of vials. The septa were fine. The vials were fine. The caps were fine.

What identified it: we measured applied torque on 30 vials from the retained sample. The mean was 1.35 N·m with a standard deviation of 0.18 N·m. The upper spec is 1.2 N·m. The capping station had drifted high, and the distribution was wide enough that roughly 20% of the lot was above the upper limit. The high torque deformed the septum, and the first needle puncture coring through the deformed face.

The fix: recalibrated the capping station to a target of 1.05 N·m with a control limit of ±0.10 N·m, and added a torque check to the outgoing record as a measured value, not a pass/fail. The coring rate dropped below 0.1%. The lesson: torque is a distribution, not a setpoint. The outgoing record has to carry the mean and the spread, not just “within spec.”

The sampling plan is not the same as the process control plan

Outgoing inspection is a gate, not a control loop. It tells you whether a lot is acceptable. It does not tell you whether the process is stable. Those are different questions, and conflating them is how suppliers end up with a plan that passes every lot and still ships bad product.

The outgoing plan should be conservative on critical defects and light on minor defects. The in-process plan should be the opposite: heavy on dimensional and torque measurement, because that is where drift shows up first. If you are running a full AQL sample on minor cosmetic defects at the outgoing gate and a spot check on torque at the capping station, you have the emphasis backwards.

For the incoming side, the customer’s own receiving inspection is a separate plan with its own AQL. That is outside this article. What matters here is that the outgoing record has to be complete enough that the customer’s receiving inspection can be a verification, not a re-test. If the customer has to re-run the leak test because your record only says “pass,” you have shifted cost onto them and they will notice.

A decision checklist for setting up the plan

When I set up an outgoing plan for a new vial SKU, I work through this list in order:

  • Define the defect classes against the test method. Critical = leak, puncture, torque out of range, wrong material. Major = dimensional out of tolerance, visible inclusion. Minor = cosmetic. Write the photo standard for minor.
  • Set the AQLs. 0.65 critical, 1.5 major, 4.0 minor is the default. Tighten critical to 0.25 only when the downstream method justifies it.
  • Pick the inspection level. General Level II unless the customer’s PO says otherwise. Get the code letter in writing.
  • Build the sample pull procedure. Stratify across pallets and cartons. Do not pull from the top carton. Document the pull.
  • Write the record form. Measured values, defect log, disposition, inspector name, date. No “pass” without a number.
  • Define the reject and rework path. A rejected lot gets a rework record and a full re-inspection, not a spot check.
  • Set the trend review. Review the defect log across the last 20 lots every month. A rising minor count is the early warning for a major defect six weeks out.

The dimensional references for the vial body and neck come from the lab vials catalog drawings, and the closure dimensions from the GPI thread standard. The GPI 9 mm thread is the one to check against for screw-thread vials, and the 11 mm crimp neck for crimp vials. If the drawing and the gauge disagree, the drawing wins, and the gauge gets recalibrated.

For labs running regulated methods, the outgoing record should reference the applicable standard. USP <661> covers the glass container, and EPA SW-846 covers the container requirements for waste and materials testing. If the vial is going into a SW-846 method, the outgoing record should carry the container clause reference. That is a one-line addition that saves a customer audit finding.

Where the plan breaks down in practice, and how to fix it

Three failure modes show up over and over in outgoing inspection schemes for vials:

The sample is not random. The inspector pulls from the carton at the front of the pallet because it is easy to reach. That carton was packed first, from the beginning of the molding run, when the mold was cold. The sample is biased toward the worst part of the lot, or the best, depending on the run. Fix: a written pull map that assigns each sample unit to a specific carton and pallet position.

The record is a checkbox. The form has “leak test: pass/fail” and the inspector checks pass. No numbers. Fix: replace every pass/fail field with a measured value field and a spec limit. If the inspector cannot write a number, the test did not happen.

The AQL is set by the customer’s PO and never reviewed. A customer specifies AQL 4.0 on critical because that is what their template says. Four percent critical defect rate on a vial that leaks is not a plan, it is a liability. Fix: push back. Show the customer the leak rate that AQL 4.0 implies and let them decide. Most will move to 0.65 once they see the arithmetic.

One more: the plan is not linked to the reorder or the shipment release. A lot that fails outgoing inspection should not be releasable by a supervisor override. The override is the single most common path for a bad lot to reach a customer. If you need an override, it goes through the quality manager, in writing, with the deviation documented and the customer notified. No exceptions.

For suppliers running multiple vial SKUs, the plan should be a single document with a SKU-specific appendix, not a separate plan per product. The AQLs and defect classes are the same across SKUs. The test methods and spec limits differ. One document, one training record, one audit trail.

The decision rule

Set critical defects at AQL 0.65 with a pressure-decay leak test at 0.6 bar and a torque measurement at 0.9 to 1.2 N·m, sample at General Inspection Level II with the code letter written into the record, and require a measured value for every critical and major test. If the record has a checkbox where a number should be, the lot does not ship. That single rule, applied consistently, is what separates an outgoing inspection plan from a stack of signed forms. The Aijiren line and the rest of the catalog run against this rule, and the customers who audit us see the same fields on every lot. If your supplier’s record does not have them, ask why. The answer is usually that the plan was never built, only inherited.

References & further reading

Related resources from our lab consumables network:

External references and standards: