A 50,000-vial lot of 9 mm screw-thread vials left the line at 12.4 mm body OD with a mean septum puncture force of 4.1 N, and the outgoing inspection record still cleared it. Four weeks later a customer running a 96-well autosampler flagged 11 coring events in 2,000 injections. The measurement that explained the complaint was not the mean puncture force. It was the lower 3-sigma limit on the 30-vial pull sample, which sat at 2.6 N, below the 3.0 N floor the customer’s needle geometry needed. The AQL sampling plan never asked for that number. That is the failure mode most outgoing QC programs have: they sample the right quantity and measure the wrong attribute.
This is a working guide to an HPLC vial AQL sampling plan and outgoing QC that actually catches the failures a customer will see. It assumes you already run in-process controls at the forming line and that incoming inspection at the lab is somebody else’s article. The scope here is the gap between “the lot is built” and “the lot ships.”
Most suppliers I have audited run a single AQL across the whole vial. That is the wrong unit of analysis. A vial has at least four independent failure surfaces, and they do not carry the same consequence at the customer’s bench. Classify first, then assign the level.
Critical, major, and minor, defined by what happens at the autosampler
A critical defect is one that can stop a run or invalidate a result without warning. In practice that is a dimensional excursion that causes needle collision or vial hang-up in the tray, a septum that is not seated (falls out under needle pressure), or a cap that will not hold torque. Critical defects get AQL 0.065 or tighter, and many buyers write zero acceptance on the critical list regardless of AQL. I would not argue with them.
A major defect degrades performance but does not stop the run: septum durometer out of band, thread dimensions that torque but strip at the high end, visible glass defects in the injection zone, dimensional drift on body OD beyond the drawing. Major gets AQL 0.65 to 1.0.
A minor defect is cosmetic or handling-related: label placement, minor scuffs outside the optical path, packaging print. Minor gets AQL 2.5 to 4.0. Nobody should be paying inspection labor to chase these.
| Defect class | Typical examples (HPLC vial) | AQL | Inspection level | Accept / reject at n=125 |
|---|---|---|---|---|
| Critical | Neck ID out of spec (needle collision), septum unseated, cap will not hold 15 in·oz, body OD > 12.65 mm | 0.065 | General II | 0 / 1 |
| Major | Septum durometer ±5 Shore A, thread pitch drift, glass defect in optical path, dimensional drift > 0.15 mm | 0.65 | General II | 1 / 2 |
| Minor | Label misalignment, cosmetic scuff outside optical zone, carton print | 2.5 | General II | 7 / 8 |
These accept/reject numbers come from ANSI/ASQ Z1.4 (the successor to MIL-STD-105E), general inspection level II, single sampling, normal severity. If you are buying to ISO 2859-1 instead, the tables are harmonized at these levels. Use one, not both, and put the standard and the level in the purchase order so the supplier cannot quietly downgrade to level I when the schedule tightens.
Sample size follows lot size, and lot size should follow the shift
At a 50,000-vial lot and general level II, the code letter is N and the sample size is 500. At 3,500 vials, code letter H, sample size 50. The math is not the hard part. The hard part is deciding what a “lot” is. I have seen suppliers batch 200,000 vials from three shifts and four mold cavities into one lot number, then draw 500 pieces and call the lot homogeneous. It is not. If a lot spans more than one shift, one glass forming cavity, or one septum molding batch, split it. The cost of an extra 50-piece pull is trivial next to the cost of a recall.
Attribute sampling alone will miss the failures that matter
Go/no-go gauging catches a vial that is out of spec. It does not catch a lot that is drifting toward the spec limit with a distribution that will not hold at the customer’s autosampler. That is the gap where most outgoing complaints live. The AQL plan needs a variables component alongside the attribute plan.
What to measure as a variable, and at what sample size
For an HPLC vial in 12 x 32 mm format, the variables I want on every outgoing lot, at n=30 minimum:
- Body OD at three heights, target 12.00 mm ±0.20 mm per the drawing, reported as mean ± 3σ
- Neck ID, which drives needle clearance, target per drawing, reported as mean ± 3σ
- Height (32.0 mm nominal), because autosampler gripper geometry does not tolerate a 0.5 mm stack
- Septum durometer on a 10-piece subsample, Shore A
- Septum puncture force on a 10-piece subsample, reported as the lower 3σ limit, not just the mean
- Cap removal torque at the high end of the crimp or screw spec, in in·oz
That last one is where I have seen the most arguments. A screw cap that torques to 15 in·oz on the assembly line but strips at 12 in·oz in the customer’s hand is a major defect that a go/no-go gauge will pass every time. Measure the stripping torque on a 10-piece subsample, not the seating torque.
The lower 3σ rule on puncture force
Puncture force data on septa is roughly normal with a fat left tail. A mean of 4.1 N with σ of 0.5 N gives a lower 3σ of 2.6 N, which is below the 3.0 N floor most 22-gauge autosampler needles need for reliable penetration. The lot passes the mean. It fails the tail. If your outgoing record only reports the mean, you are shipping the tail to the customer. Report the lower 3σ and set the acceptance limit on that number. This is the single change I would make first in any outgoing QC program.
What a realistic outgoing inspection record actually contains
I have reviewed a lot of CoAs. The good ones share a structure. The bad ones are a single page with a green checkmark and a lot number. Here is what I want to see, and what I have argued with suppliers about.
The header: lot identity, drawing revision, and the standard invoked
The record should open with the lot number, the manufacturing date, the drawing revision the lot was built to, the standard invoked (ANSI/ASQ Z1.4 or ISO 2859-1), the inspection level, the AQL per defect class, and the sample size actually drawn. If the drawing revision is not on the record, the record is not traceable. I have seen a customer reject a shipment because the record cited a drawing revision that predated a neck geometry change. The vials were fine. The paperwork was not.
The attribute section: defects found, not just pass/fail
List each defect class, the number of defects found in the sample, the acceptance number, and the disposition. If the sample found 0 critical, 1 major, and 3 minor against acceptance numbers of 0/1/7, the record should say so. A record that says “PASS” without the counts gives the customer nothing to trend. Trend data is what lets a customer reduce incoming inspection, which is the outcome your sales team wants.
The variables section: mean, sigma, and the limit
For each variable, report n, mean, standard deviation, and the lower or upper 3σ limit alongside the specification. A line like “Body OD: n=30, mean 12.02 mm, σ 0.04 mm, upper 3σ 12.14 mm, spec 12.00 ±0.20 mm” tells the customer the lot is centered and the tail is inside. A line like “Body OD: 12.02 mm” tells them nothing about the tail. Push for the full line.
The functional section: the tests that mimic the customer’s bench
This is where a serious outgoing record separates from a checkbox. Include:
- Septum puncture count to failure on a 10-piece subsample, reported as median and minimum. A 9 mm silicone/PTFE septum should survive 50+ punctures at 22-gauge before coring. If the median is 30, the lot is suspect even if the mean force is good.
- Extractable screen on a 5-piece composite, typically a blank injection on a C18 at 254 nm with a 10 µL injection, reported as the largest peak area above a stated threshold. This is the test that catches a septum polymer change the supplier did not disclose. If you only run one functional test, run this one.
- Cap torque retention after 24 hours at 40 °C, reported as the percentage of initial torque retained. This catches the caps that loosen in a hot shipping container.
The traceability block and the signature
Lot number, glass batch, septum molding batch, cap batch, and the inspector ID with a date. If any of those links is missing, the customer cannot do a lot-to-lot comparison when a problem appears, and neither can you. The chain has to be complete enough that a complaint can be traced back to the molding or the batch. I have written elsewhere about the five-link traceability chain; the outgoing record is where it either exists or does not.
Two field cases where the sampling plan, not the product, was the problem
Case 1: the drift that passed every attribute check
Symptom: a customer running 2 mL vials on a well-known autosampler reported intermittent needle misalignment, roughly 1 in 4,000 injections, always on the same tray position pattern. Measurement: the outgoing record showed body OD at 12.03 mm mean, well inside 12.00 ±0.20 mm. But the record had no σ. When we pulled 30 retained samples from the same lot, σ was 0.09 mm, and the upper 3σ was 12.30 mm, which is outside the spec. The lot had a bimodal distribution from two mold cavities, one of which was drifting. Fix: the attribute plan passed the lot because the sample of 500 caught only a handful above 12.20 mm. We added a variables pull of n=30 per cavity, reported the upper 3σ, and set an acceptance limit at 12.15 mm on the 3σ number. The next lot from the drifting cavity was held. The customer’s misalignment rate dropped to zero within two shipments.
Case 2: the extractable that only showed up on the composite
Symptom: a pharma customer running gradient methods at 210 nm reported a ghost peak at 6.8 minutes on roughly 1 in 20 injections, always from a fresh box. Measurement: the outgoing record showed all attribute checks passed and puncture force was nominal. But the extractable screen was run on individual vials, not a composite, and the threshold was set too high to catch a small peak. When we re-ran the screen on a 5-vial composite at 210 nm with a 20 µL injection, the peak was there at 0.8 mAU, well above the 0.2 mAU threshold the customer needed. It traced to a septum molding batch that had switched to a different peroxide cure. Fix: the extractable screen moved to 210 nm, composite sampling, threshold 0.2 mAU, and the septum molding batch was added to the traceability block. The ghost peak did not recur on the next 14 lots.
Both cases share a pattern. The attribute plan was doing its job. The variables and functional sections were not. If your outgoing record is all attributes, you are shipping the tail of a distribution to a customer who will find it.
Where the sampling plan meets the purchase order
The AQL plan is only enforceable if it is in the PO. I have seen suppliers quietly move from level II to level I when a schedule tightened, which cuts sample size from 500 to 200 at the same lot size and roughly doubles the customer’s risk of accepting a bad lot. The PO should name the standard, the inspection level, the AQL per defect class, the variables to be reported, the functional tests, and the disposition rule on a rejected lot. If the supplier cannot meet the plan, they should say so before the lot ships, not after.
For the customer side, the same logic applies in reverse. If you are buying from a distributor rather than a manufacturer, ask for the outgoing record from the actual manufacturing lot, not a distributor-level inspection. The HPLC vial you receive should carry a record that traces to the molding batch, not just to the distributor’s warehouse. A distributor-level CoA that does not name the glass batch or the septum molding batch is a marketing document.
The sampling plan should be sized to the customer’s risk, not the supplier’s convenience
Here is the decision rule I would put in front of any procurement manager writing a vial specification. Size the plan to the defect class and the consequence, not to the lot size alone. Critical defects get AQL 0.065 and zero acceptance. Major defects get AQL 0.65 and a variables pull of n=30 per cavity with the 3σ limit on the record. Minor defects get AQL 2.5 and a cosmetic check. Run the extractable screen at the customer’s detection wavelength on a composite, because that is the test that catches the failures the attribute plan cannot see. Put the standard, the level, and the reporting format in the PO, and require the traceability block to name the glass batch and the septum molding batch. If a supplier cannot produce that record, the sampling plan is not the problem. The supplier is.
For the broader consumables context around this, the chromatography consumables catalog and the lab vials reference pages are useful for comparing formats, and the Aijiren brand documentation is one place to see what a full outgoing record looks like when it is done properly. For method-level context on what the customer will do with the vial, the U.S. EPA SW-846 test methods and the Restek technical resources are worth reading alongside your own method. The Agilent chromatography consumables and columns pages and the Thermo Fisher chromatography and sample handling documentation both publish autosampler geometry that your neck ID and body OD specs need to clear. Use those numbers when you write the drawing, and use this sampling plan when you write the PO.
The decision rule: classify the defect, set the AQL by class, add a variables pull at n=30 per cavity with the 3σ limit reported, run one functional extractable screen at the customer’s wavelength, and refuse any lot whose record does not name the glass batch and the septum molding batch. That is the whole plan. Everything else is paperwork.
References & further reading
Related resources from our lab consumables network:
- Hplc Vial at vialforhplc.com
- Chromatography Consumables at chromatographyconsumables.com
- Lab Vials at labvials.com
- Aijiren Brand at aijirenvials.com
External references and standards: