A 40-lot audit of incoming 20 mL headspace vials last spring turned up 11 lots with septum puncture force above 9.5 N, against a specification of 6.0 to 8.0 N. Every one of those lots came from a supplier whose quality manual listed a durometer check on the incoming compound but no lot-level release criterion. The vials passed dimensional inspection, passed visual, passed the certificate of analysis. They failed in the autosampler at puncture 3 or 4 because the needle was coring the PTFE face. That is the shape of the problem a voa vial supplier audit checklist has to catch: not whether the vendor has a quality system, but whether the system has a measurable release gate at the step that determines the failure mode.
Most audit templates I’ve seen are built for machined parts. They ask about calibration intervals, document control, nonconforming material segregation. Useful, but they don’t tell you whether the vendor can hold a ±0.20 mm body OD on a molded glass article, or whether their septum supplier is actually the one named on the CoA. This article is the checklist I use now, built from the specific failure modes of volatile organic analysis containers and the evidence that predicts them.
Every supplier will send a CoA. The question is whether the CoA reports a value against a limit or a value against nothing. I request the internal specification document for the finished vial, not the customer-facing CoA, and I read it for three things: are there numeric limits, are they tied to a test method, and is there a sampling plan.
A specification that says “septum: silicone/PTFE, suitable for GC” is not a specification. A specification that says “septum: 45 ± 3 Shore A silicone, 0.010 in PTFE face, puncture force 6.0 to 8.0 N per ASTM D3767 on lot sample n=32 per ISO 2859-1 general inspection level II, AQL 0.65” is. The difference is whether the vendor can reject a lot before it ships. Vendors without numeric limits ship whatever the molding machine produced that shift.
The five release criteria that actually predict failure
From my own failure history, these five items account for most of the field problems I’ve traced back to supplier control:
- Body OD and concentricity. Nominal 22.50 mm for a 20 mL vial, tolerance typically ±0.20 mm, concentricity within 0.15 mm TIR. Out-of-round bodies jam autosampler grippers and cause the needle to enter off-center, which is where coring starts.
- Neck ID and thread form. The GPI 18-400 finish has a defined thread pitch and root diameter. A neck that mics within tolerance but has a thread root that’s 0.10 mm shallow will not seal at rated torque. Ask for the thread gauge drawing.
- Septum puncture force and reseal. Puncture force 6.0 to 8.0 N, reseal tested by holding 10 psi for 30 seconds after 3 punctures with a 23-gauge needle. Vendors who don’t test reseal after multiple punctures are the ones whose septa leak on the third injection.
- PTFE face thickness and bond. 0.008 to 0.012 in, bonded, no delamination after 3 punctures. A thin or poorly bonded face is the source of the PTFE shards that show up as ghost peaks and, occasionally, as a blocked needle.
- Hydrolytic resistance. Per USP <660> and the ISO 4802 series, with the powder test for the glass type and the surface test on the finished article. This is the criterion that separates a borosilicate 33 expansion vial from a soda-lime one that was relabeled.
If a vendor can’t produce a document showing numeric limits against a test method for all five, the audit is effectively over. You are not auditing a quality system; you are auditing a hope.
Trace the septum to its compound lot, not to its part number
Septum is where the voa vial fails in the field, and it’s where supplier traceability is weakest. The vial molder buys septa from a converter, the converter buys compound from a silicone house, and the compound house buys filler and cure catalyst from chemical distributors. Any of those steps can drift without the part number changing.
I ask for the compound lot number on the septum and the cure date. Silicone compound has a shelf life; a converter sitting on 18-month-old stock will produce septa with different compression set than fresh compound, even at the same durometer. The test is compression set per ASTM D395 Method B, 22 hours at 175 °C, target under 25%. A lot at 35% will not reseal reliably after the second puncture.
What the septum CoA should contain
A usable septum CoA carries: compound lot, cure date, durometer (45 ± 3 Shore A), specific gravity, compression set, and the PTFE film lot with its thickness. If the vendor’s CoA lists only “silicone/PTFE, 45 Shore A” with no lot numbers, the traceability chain stops at the part number and you have no way to investigate a field failure. I’ve seen this exact gap turn a 3-day root cause investigation into a 6-week one.
For septa specifically, the suppliers at septacaps.com and vialsepta.com publish compound-level data more consistently than most; use their data sheets as a benchmark for what you should be demanding from any vendor.
Score the vendor on process capability, not on certification count
ISO 9001 certificates are table stakes and predict almost nothing. What predicts is Cpk on the critical dimensions. Ask for the last 12 months of SPC data on body OD, neck ID, and septum thickness, and calculate Cpk yourself. A vendor running Cpk of 1.0 on body OD is at 3 defects per thousand, which sounds fine until you realize a 100,000-piece order has 300 out-of-spec vials mixed in, and your autosampler will find them.
My threshold: Cpk ≥ 1.33 on body OD and neck ID, Cpk ≥ 1.50 on septum thickness. Below 1.33 the process is not in control and the vendor is sorting, not manufacturing. Sorting works until the sort inspector gets tired.
The scoring table I use
Score each vendor on a 100-point scale, weighted toward the items that predict field failure. Anything under 70 is a no-go. Between 70 and 85 is a conditional approval with incoming inspection at tightened level. Above 85 is approved at normal inspection.
| Category | Weight | What earns full points |
|---|---|---|
| Release criteria with numeric limits and test methods | 20 | All five criteria documented, method cited, sampling plan per ISO 2859-1 |
| Process capability (Cpk) on critical dimensions | 20 | Cpk ≥ 1.33 on OD/neck, ≥ 1.50 on septum; 12 months data provided |
| Septum traceability to compound lot and cure date | 15 | Compound lot, cure date, compression set data on every CoA |
| Glass hydrolytic resistance testing | 15 | USP <660> and ISO 4802 surface test on finished article, lot-level |
| Change control and notification | 10 | Written notification of any material, process, or sub-supplier change, 90 days ahead |
| Complaint history and CAPA closure | 10 | Last 24 months complaints, root cause, corrective action verified effective |
| Incoming inspection at vendor | 5 | AQL 0.65 at general level II, records available for review |
| Retained samples | 5 | Retention of every lot for 24 months minimum, retrievable on request |
Two things about this table. First, the weights are deliberate: 70 of 100 points sit on things you can measure from documents, not things the vendor tells you on a tour. Second, the change control item is worth more than it looks. A supplier who changes septum compound without telling you will pass every other item and still wreck your method.
Request the evidence package and read it before the audit
The audit itself is a conversation. The evidence package is where you find out whether the conversation is worth having. I ask for the following 30 days before the visit, and I read it in this order:
- Finished-goods specification for the vial, with test methods and sampling plan.
- 12 months of SPC data on body OD, neck ID, septum thickness, and puncture force.
- Last 24 months of complaints with root cause and CAPA closure evidence.
- Septum supplier list with compound lot traceability examples from the last 5 shipments.
- Hydrolytic resistance test reports at lot level, not the annual type test.
- Change control log for the last 24 months, including sub-supplier changes.
- Retained sample policy and a sample retrieval exercise (pick a lot from 18 months ago, ask them to produce the retained sample).
The retained sample exercise is the one that separates real systems from paper systems. About a third of the vendors I’ve asked cannot produce a retained sample from 18 months ago on request. That tells you the retention policy is a paragraph, not a practice.
Field case: the lot that drifted after a compound change
One of my customers had a headspace run that started showing a 0.4-minute shift in retention time on the early-eluting compounds, and a new peak at 1.1 minutes that wasn’t in the method. They chased the column first, replaced it, no change. They chased the carrier gas, no change. The measurement that identified the problem was a septum puncture force check on the current lot versus the previous lot: 7.2 N versus 8.9 N. The new lot was softer, and the softer compound was off-gassing a low-level siloxane that showed up as the 1.1-minute peak. Root cause was a compound change at the septum converter that never reached the vial molder’s change control log. Fix was a supplier change control requirement with 90-day notification and a lot-level puncture force check at incoming. The retention time shift went away with the next lot.
Verify the glass, not the label
Borosilicate 33 expansion glass has a coefficient of thermal expansion around 3.3 × 10⁻⁶ /°C. Soda-lime is roughly 9.0. The difference matters because a vial that’s mislabeled soda-lime will survive autoclaving and then fail in a heated headspace oven at 80 °C over a 30-minute cycle, usually by cracking at the neck-to-body transition. The failure is intermittent, which is what makes it expensive.
The evidence to request is the hydrolytic resistance test at lot level, per USP <660> and ISO 4802. Powder test for the glass type, surface test on the finished article. A vendor who sends only the annual type test is not testing the glass that’s going into your vials. I also ask for the glass supplier’s lot certificates and the conversion station’s incoming inspection records, because the conversion station is where a mislabeled glass lot becomes a mislabeled vial lot.
The functional test that catches mislabeled glass without a lab: heat 20 vials from the lot to 90 °C for 60 minutes, cool to room temperature, inspect at 10x for cracks at the neck transition. Soda-lime lots show a 1 to 3% failure rate; borosilicate lots show zero. It’s a crude test, but it’s fast and it’s decisive enough to reject a lot before it enters the warehouse.
Score the vendor objectively, then hold the score
The scoring table above is worth nothing if you don’t apply it consistently. I score every vendor on the same table, with the same weights, and I keep the scores in a spreadsheet that goes to the quarterly review. The vendor’s score is reviewed every 12 months, and any vendor who drops below 70 goes to conditional approval with tightened incoming inspection until they climb back.
The incoming inspection plan should be tied to the score, not to a fixed AQL. A vendor at 90 gets general inspection level II, AQL 0.65 on critical defects. A vendor at 75 gets level III, AQL 0.40. A vendor at 65 gets 100% inspection on the five critical criteria until they fix the process or lose the business. This is how you make the score mean something.
Field case: the CoA that reported a value against no limit
A second customer had a septum lot that failed reseal after 3 punctures in their leak test, showing pressure drop of 2.5 psi over 30 seconds against a 1.0 psi limit. The vendor’s CoA reported “puncture force: 7.4 N” with no limit. When we asked for the internal specification, there was none for reseal after multiple punctures; the vendor tested only single-puncture reseal. The measurement that identified the problem was the customer’s own leak test, which was more stringent than the vendor’s. The fix was to add reseal-after-3-punctures to the purchase specification at 10 psi for 30 seconds, and to require the vendor to report the value against that limit on every CoA. Two lots were rejected on the next shipment, which is exactly what the specification was for.
This is the pattern I see most often: the vendor’s test is less stringent than the failure mode. The audit checklist has to compare the vendor’s test method against the failure mode, not against a generic standard.
What to do when the vendor refuses to provide the evidence
Some vendors will refuse to provide SPC data, complaint history, or septum compound traceability. The refusal is itself a data point. In my experience, a vendor who refuses to provide 12 months of SPC data on body OD either doesn’t have it or knows it will show Cpk below 1.0. Either way, you’ve learned what you needed to learn.
The decision rule: any vendor who cannot provide four of the seven evidence items within 30 days is not approved, regardless of price. The cost of a field failure, one bad lot in a 500-sample batch, is typically 10 to 20 times the unit price difference between an approved and an unapproved vendor. The math is not close.
For vendors who provide the evidence but score between 70 and 85, approve conditionally with tightened incoming inspection and a 6-month re-score. For vendors above 85, approve at normal inspection and re-score annually. The score is the contract, in effect: it determines how much you trust the CoA and how much you verify yourself.
The final takeaway is a single rule: approve a voa vial vendor on the strength of their release criteria and their Cpk, not on their certificate or their price. If the release criteria don’t have numeric limits against a test method, and the Cpk on body OD isn’t at least 1.33, the vendor is not in control of the process that produces your vials, and no amount of incoming inspection will fix that. Audit the specification, score the capability, and hold the score. The rest is paperwork.
References & further reading
Related resources from our lab consumables network:
External references and standards: