What Certification Evidence Should Be in a Custom Battery Pack File?
A custom battery pack file should contain six distinct evidence types, and each one has to be tied by part number, revision, and manufacturing site to the exact pack named on the purchase order: a UN 38.3 test summary that lists both the cell model and the pack, a UL 2054 or UL 62133-2 certificate whose scope covers the assembled pack and not only the cells inside it, an IEC 62133-2 test report where the end-product certification path calls for one, a QMS certificate that is unexpired and shows a live surveillance cycle, a traceability record that runs from cell lot code to pack serial number, and a change-control commitment that obligates the supplier to notify before any change to cell, BMS, firmware, or site. Anything short of that set is a pile of PDFs. The register that holds them is what turns it into a file.
I learned to think this way the afternoon I opened a supplier's "certification package" with the PO in one window and the supplier's cover note in the other. Nine attachments. One of them was a scan rotated ninety degrees. The file names were UN38.3.pdf, UL cert.pdf, ISO9001.pdf, MSDS.pdf, and a folder called new folder (2). Nothing in the email said which pack any of it belonged to.
The exercise that afternoon took two hours and almost none of it was reading. It was matching.
The documents don't make the same promise
A test summary is not a test report, and a certificate is not either of them. The summary is a short table: who tested, against what standard, which cells, which tests were run, pass or fail. It exists because the transport rules require a summary to move with lithium cells and batteries. It is a statement that a test happened. It is not a statement that a product is certified for the market you're selling into.
The report is the document underneath the summary. It has the data pages, the sample description, the photos of the unit under test, and the deviation notes that tell you what the lab actually did that day. When a supplier sends a summary and says the report is proprietary, you are being asked to buy on a three-page table with no data behind it. Sometimes that's genuinely all that exists. Cell makers frequently release the summary and hold the report, because the report was written for the cell, not for your pack. When that happens, I want the pack-level certificate carrying the weight instead, and I want the supplier to state in writing that the summary belongs to the exact cell model on the PO. I have seen one summary circulate across three different cell models with the same revision number, because nobody edited it after the first test round.
A certificate is a third-party attestation. It carries a number, an issue date, an expiry, a scope statement, and usually a list of the models or constructions it covers. A declaration of conformity is the manufacturer's own signature. Useful, but it's a claim, not an attestation. A QMS certificate says the factory runs a controlled process. It says nothing about whether your pack passed anything. I sat in a review once where an ISO 9001 certificate was accepted as the safety evidence for the pack, and I understand the instinct — it was the only document in the folder with a logo and a number on it. It still didn't answer the question anybody was asking.
The part number is the hinge
Every document in the file has to be tied to the same four identifiers: the cell model, the pack part number and revision, the BMS firmware revision, and the manufacturing site. I keep those four in a header row at the top of the register and check each document against them, line by line, before I read anything else. It sounds mechanical. It is mechanical. That's the point.
The one that catches people is the scope of a pack certificate. A certificate that covers cells does not automatically cover the assembled pack unless the pack is specifically listed or tested as a pack. Sometimes the certification path runs through a pack-level listing. Sometimes it runs through a listed cell used inside its ratings, with the pack covered through the end product's own evaluation. Which one applies depends on the end product, the sales channel, and how the certifier built the file. What does not depend on anything is this: you have to be able to point at the line in the document that names your pack. If the only place your part number appears is the supplier's cover email, the certificate is not for you.
I also check the address on the QMS certificate against the address of the plant in the PO. Factories get folded into a sister plant's certificate during busy stretches. The certificate is real, the quality system is real, and the pack you're buying is being built somewhere the auditor never walked.
Checking the file against something other than itself
A file can be internally consistent and still wrong. The check that catches it takes about ten minutes: take the certificate number and look it up in the issuing body's public registry. What you want to see is the scope, the expiry date, and whether the surveillance audits are current. A withdrawn certificate looks perfectly fine as a PDF. Surveillance status is the part people skip — a certificate can sit inside its expiry window and still be suspended, and the registry is where that shows up.
The other outside check is the traceability sample. Ask the supplier to pick a pack serial from last month's shipment and walk it back to the cell lot code, the incoming inspection record for that lot, and the date the pack was built. A supplier with real traceability does this in a day. A supplier without it sends a process description, which is a different thing. I have never watched a supplier who was already good at this fail the test.
The register, and the column that keeps it honest
The register is a flat table. Document type, revision or certificate number, issue date, issuing lab or body, the part number it applies to, and the next review date.
The last column is the one that matters and the one everybody leaves out.
Here's why. Certificates expire. QMS surveillance lapses. Cell models get discontinued and replaced with something the supplier swears is identical. Firmware gets updated to fix a field problem, and depending on how the pack was certified, that update can move the product outside the tested configuration. None of it announces itself. The register is a drift detector. If the next review date on the UN 38.3 summary for the cell in your highest-volume pack lands six weeks out, you find out now, with room to act, instead of in a loading dock when somebody asks for the paperwork before the truck leaves.
Red flags I treat as hard stops rather than questions: a test summary with no part number anywhere on it, a report delivered without the data pages, a certificate naming a factory you've never audited, and a QMS certificate that expired two months ago while the renewal is "with the certifier." Each of those is a document that will not survive a customer's compliance review, and the time to fix it is while there's still a PO to hold back.
Where it depends
Not every pack needs the whole set. A pack that ships as a component into a listed end product has a different evidence path than one sold as a standalone accessory, and a supplier building a few hundred industrial units a year may honestly not carry a pack-level listing because the volume never justified it. Retailer and marketplace compliance programs pull harder on some documents than federal rules do, and they change their lists on their own schedule. I've had to go back for a document a year after the file was signed off because a customer's onboarding checklist grew a line.
What doesn't vary is the method. Decide which evidence the path actually requires, write those into the supply agreement, tie every piece to the four identifiers, verify the third-party ones at the source, and keep a review date on all of it. The file is not a deliverable you receive once and shelve. It's a document you maintain for as long as you buy the pack.
It reads like a procedure when I write it out. In practice it's a spreadsheet with an autofilter and about forty minutes a quarter, and it is the cheapest insurance in the entire pack sourcing process.
