Who holds the pack certificate when cells and assembly come from different suppliers?

Certification ownership follows the entity named on the document, not the entity that built the product. In a split supply chain, the cell maker holds the cell-level evidence (UN 38.3 test reports, IEC 62133-2 cell certificates), the pack assembler holds the pack-level listing or certificate for the configuration it actually builds — usually under its own model number — the BMS developer holds approvals that stop at the board, and market-entry approvals sit with whoever places the finished pack on the market, which in most OEM programs is the OEM. If the RFQ does not name a holder for each document, the default outcome is that nobody holds it, and the buyer owns the gap.

The lesson arrived from a freight forwarder's hazmat desk, not from a supplier audit. Their standing question for anything with lithium in it is short: send the UN 38.3 test summary, and make sure the model number on it matches the model number on the carton. So I sent the file the assembler had handed me at the end of the build.

An hour later it came back with one line: the header says PK-4813-B, the cartons say our part number, which one is on the certificate?

Both, technically. The assembler's test summary covered the standard enclosure they build and sell under their own number. Our unit was that enclosure with our label, our BMS, our firmware, and our model number on the carton. The testing was real. The T1 through T8 table was complete and every line passed. It simply did not name our product, and the person on the dock in Memphis does not care about the nuance. He cares that the paper and the box match.

Nobody had lied to me. That is the part I want you to sit with, because it is the part the generic advice misses. Everyone in the chain did their job correctly and told the truth. The cell maker had cell reports. The assembler had a pack listing. The BMS house had approvals for its board. Each party assumed the next party up the chain held the thing that covered the finished article, and the finished article was ours.

The name on the paper is the whole argument

I stopped thinking about this as a product question and started thinking about it as a document question. Every certificate in the chain has a field that says who it belongs to. In a UL listing you will see the listee — the company holding the file — and often a separate manufacturer field and a factory field. On an IECEE CB test certificate you get an applicant and a manufacturer, and they are frequently not the same company. On a UN 38.3 test summary, the name is whoever commissioned the testing, because the summary is not a certificate at all: it is a condensed version of a laboratory report, issued to the client who paid for it.

That last point matters more than people expect. A test summary naming the assembler is often perfectly usable by you, because a forwarder's real concern is that the summary matches the model number and the manufacturer shown on the shipping documents. The mismatch stops the shipment. The ownership only becomes the problem when you need the document to travel with your name on it, or when you need it reissued because something changed.

So the first thing I do now, on any split program, is build a register before the first PO. One row per document. Columns for the document, the holder of record, where the physical file lives, the revision, what triggers a reissue, and who is obligated to tell whom when that trigger fires. Cell reports. Pack listing. BMS EMC approval. Firmware safety evaluation. Market-entry approvals for the finished product. Nothing sophisticated — a spreadsheet will do — but it exists on day one instead of the day a customer's supplier portal asks for it.

Where the gap actually opens

It opens at the model number, and it opens again at the cell.

An assembler can only certify the configuration it builds. If they build ten thousand of their standard pack and you take two thousand with your own label and your own harness, their certificate describes their artifact. Getting your name into that file is possible in some schemes — the OEM can hold its own listing, or the parties can arrange for your name to appear — but that is a setup task that has to happen before production, and it has a cost and a timeline attached. Ask about it in the RFQ, not in month nine.

The cell is the quieter trap. Pack listings are frequently issued against a specific cell from a specific maker. Your assembler qualified a particular 21700, the listing names it, and then eighteen months into the program the cell maker sends an end-of-life notice and offers a successor. The successor may be electrically identical on paper. It is still a different cell, and the listing that names the old one now describes something you no longer build. This is also where the third-party BMS bites: a board-level approval covers the board, and the pack-level safety evaluation still has to cover the combination — cells, board, housing, thermal path, and the firmware that decides when to open the contactor. A firmware change that moves a protection threshold can invalidate the evaluation that covers the assembly. Nobody in the chain will volunteer that, because from their seat nothing changed.

There is a second-source problem hiding here too. If the pack certificate is held in the assembler's name, and two years from now you want a second assembler building the same design, you cannot carry that certificate across the street with you. You are starting the pack-level evaluation again, on a timeline you did not plan for, in front of a customer who thinks the product is already qualified.

What goes in the RFQ

I ask every bidder three questions per document, and I ask them in writing so the answers are on the record:

• Do you hold this document today, and can you send the current revision with the bid?
• Will you obtain it under this PO, and on what schedule?
• Whose legal name will appear on it as the holder?

That third question is the one that changes conversations. Most bidders answer the first two without thinking. The third one makes them go look. If the answer is anything other than your company's legal name, you now know exactly which documents you are relying on someone else to keep alive, and you can decide whether that is acceptable.

I also ask two things that are not about certificates at all. First, how the bidder notifies you of design changes — a cell substitution, a BMS firmware release, a housing revision, a change in the factory that molds the case. Second, what happens to the evidence if you leave. Not the tooling, the evidence. Who keeps the test reports, for how long, and in what form, if the relationship ends.

The quality agreement is where this lives, not the PO

The PO is the wrong instrument for this. It is transactional by design and it gets superseded every time you reorder. What I want is a quality agreement that records, in plain terms, which party maintains which piece of evidence, how long it is retained, in what format, and how changes are notified between the parties. Retention needs to be a number of years, not the phrase "as long as necessary," because "as long as necessary" gets interpreted by whoever is answering the question that day.

I do not draft the language that allocates responsibility or liability. That is not my job and it is not the buyer's job. What I do is hand counsel a one-page list: here are the documents, here is who we believe holds each one today, here is who we need to hold it, and here are the events that should trigger a reissue. Counsel turns that into an obligation with teeth. If I try to write the clause myself, I get a clause that sounds right and allocates nothing.

One more habit, and it is the cheapest one. Every time a carrier, a customer's supplier quality portal, or a customs broker asks me for a document, I write down which document they asked for and which party had it. Nine times out of ten the answer is fine. The tenth time is the one that tells you where your register has a hole, and it is much better to find that hole on a routine request than during a recall conversation or a shipment sitting on a dock.

The reason this topic keeps coming back is that a split supply chain is genuinely good at building products and genuinely bad at remembering who owns the paper. The cell maker is not going to volunteer that their report does not cover your pack. The assembler is not going to volunteer that their certificate names their model. The BMS developer is not going to volunteer that the safety evaluation has to cover the assembly, not the board. None of that is dishonesty. It is just the shape of the chain. Your job, sitting at the point where all of it converges, is to make the RFQ say out loud which name goes on which document — before anyone starts building.