What a Compliance Reviewer Checks in a UN 38.3 Test Summary

What a Compliance Reviewer Checks in a UN 38.3 Test Summary

A UN 38.3 test summary is a procurement document that describes one specific battery configuration, not a certificate that covers every pack a supplier builds under a part number family. A reviewer reads it to answer a single question: is the battery on my purchase order the battery that was tested? If the answer isn't a clean yes, the summary isn't evidence — it's a question nobody has asked yet. The document belongs in your hands with the quote, reconciled against your own pack spec before a purchase order exists, because a configuration mismatch is a design decision, not a paperwork problem.

Come sit at the desk with me. I'll read one out loud.

Three windows and a highlighter

I read these with three things open. The summary PDF sits on the left half of the screen. The supplier's datasheet for the exact part number I intend to buy sits on the right. Behind both is my own pack spec — the one that came out of the robot's duty cycle, the voltage window, the connector pinout, and the mounting envelope I have to fit into. There's a highlighter in my hand even though I'm reading on a screen. The habit is from paper and it hasn't died.

If I can't lay those three side by side, I'm not reading. I'm skimming, and skimming is how a substitution gets through.

I read in a fixed order: header, description, test table, edition line. That order isn't arbitrary. It's the sequence that catches what people actually get burned by, and what people get burned by is almost never a lab faking a number.

The header, where ownership sits

The first block names the manufacturer of the battery and gives contact information. Read that name slowly. Not the logo in the page footer — the name in the field.

The manufacturer of the battery is usually not the cell maker. It's the entity that assembled the pack: the one that chose the cell arrangement, the protection board, the housing, the hardware. That entity is the one that has to reissue the summary when something changes. So when the name in the field is a company I've never heard of and the quote came from a distributor, I don't panic. I just note that the OEM is the escalation path, and I make sure I have a working contact there before I need one.

Then the test laboratory: name and contact details. A lab name with no address is an open item. I write it down and keep moving. I'm not accusing anyone of anything at that point. I'm recording that I can't verify a party, and that note follows the file.

Then the unique test report identifier and its date. I copy the identifier into my tracker character for character, because eight months from now, when a customer's logistics team asks about a unit that's already on a truck, that string is the difference between finding the right PDF in ten seconds and hunting through a shared drive for twenty minutes. The date goes down next to it. I hold that date in my head while I read the next section, because in about a minute it's going to matter.

The description line is where deals actually die

This is the paragraph that names the cell, the chemistry in general terms, the arrangement, the capacity, the voltage, and the shape.

I put my finger on the cell manufacturer and model first. Then I look at the datasheet window. Same cell? Not "same cell family." Same model string. If the pack I'm buying runs a different cell than the one in the tested article, everything below this line is somebody else's test result. I stop reading and start a conversation.

Then the arrangement. Fourteen series, six parallel. Ten series, four parallel. Whatever it says has to match the pack I'm buying, because the vibration and shock results belong to a mass and a structure, not to a chemistry.

Then the watt-hour rating. That number has to agree with the datasheet, and it has to agree with what will end up on the shipping paperwork. When the summary, the datasheet, and the label on the physical unit disagree about watt-hours, nobody has lied — somebody has been sloppy — and I'm the one who reconciles it. Better now than at a dock.

The most common mismatch in this section is quiet. The part number I'm buying reads like the tested one. Same family, same prefix, one digit apart, and that digit is an extra parallel group. It looks clerical. It isn't. It's a different battery.

Eight tests, and which article each one belongs to

There are eight rows in the underlying standard: altitude, thermal, vibration, shock, external short circuit, impact or crush, overcharge, and forced discharge. What I check is not just that they're listed with results. I check which article each one was run against — the cell, the battery, or both — because they don't distribute evenly.

Altitude, thermal, vibration, shock, and external short circuit are run on cells and on batteries. Impact is a cell test; crush is the battery equivalent. Overcharge applies to batteries. Forced discharge applies to cells.

So a summary that shows all eight stamped against a single line reading "battery" tells me the person who assembled the document wasn't paying attention, or the template is sloppy, or both. That's not automatically a rejection. It's a question I ask in writing, and I read the answer carefully — because the answer tells me how the supplier handles documents in general, which is a preview of how they'll handle mine.

One more line before the table, and it's the one everybody forgets: the edition of the Manual of Tests and Criteria the testing was run against. I check two things about it. The edition cited should have been in effect when the testing was performed, and the date in the header should sit inside that window. Section 38.3 has been tightened and clarified across revisions, so a very recent edition cited against an old test date is usually a document that was edited after the fact. Sometimes that's innocent. It's always worth one question.

What the summary does not cover

Here's the part that catches people in this industry.

The purchased pack is covered only to the extent that its cell, configuration, protection electronics, and enclosure match the battery described in the summary. Four things. Miss one and the rest of the document is describing someone else's product.

The protection electronics are the one everybody skips. Overcharge and external short circuit results depend on the protection circuit that was inside the tested article. If the supplier's current build uses a revised board — even one that does the same job with different parts — that goes in writing as a question: can this change affect the outcome of any of the eight tests? If the honest answer is "we don't know," then it's a retest, and the retest is a line item in the purchase order, not a promise in an email.

Same logic for the enclosure. Housings, potting, cell holders, vent paths. Crush, impact, and thermal results live and die by those choices.

And then the one that requires the most discipline for the length of a program: has anything changed since the test date? A design change that could affect the outcome of any of the eight tests requires the battery to be retested and the summary reissued for the new configuration. That isn't a supplier preference. That's how the standard works. So the change notification commitment — cell, board, housing, anything — matters for years after the pilot, and I want it written into the agreement in plain language. Tell me before, not after.

Transport evidence is not product safety evidence

UN 38.3 is subsection 38.3 of the UN Manual of Tests and Criteria. It is a transport safety requirement. It is not a product safety certification, and a clean summary doesn't answer whether the pack is safe inside my robot, on my dock, over ten thousand cycles.

Those are different documents. UL 2054, UL 2271, IEC 62133-2 — sometimes pulled in by a customer's own standard, sometimes by an end user's procurement policy, sometimes by nobody until an integrator asks late in the program. A summary with every box checked is a floor, not a ceiling. I've watched programs treat UN 38.3 paperwork as if it closed the safety file. It closes exactly one question, and it closes it for transport.

The sequencing, which is the actual skill

The reading matters. The order you ask for the reading matters more.

At the RFQ stage, the summary request goes in the same email as the price request, tied to the exact part number. Not the family — the part number. Along with it: the test report identifier, the lab name, the test date, and one direct question. What has changed in this design since that date?

If the pack is new and hasn't been tested in my configuration, then the retest gets scoped now, dated now, and written into the purchase order as a deliverable. Not promised verbally. Not "we'll take care of it." Scoped and dated as a condition of the first shipment.

Before the order is placed, I reconcile the summary against my spec sheet and write down every open item on one page. If the configuration differs, I make a decision: does the supplier retest, or do I change the pack? Those are the only two honest options, and both of them cost something. Choosing to ignore the difference is a third option people take, and it's the one that comes back.

At pilot, I verify the delivered hardware against the tested description. First-article inspection. Cell markings on the actual units, not the datasheet. A spec sheet and a shipped unit can diverge, and the divergence is cheaper to fix at ten units than at ten thousand.

And then the part nobody warns you about. In the United States, lithium battery transport rules sit with PHMSA in 49 CFR 173.185, and the expectation is that the test summary is available on request. "On request" doesn't mean from the supplier's portal whenever it loads. It means when someone asks. And who gets asked? Often me — because I'm the one shipping returned units, warranty swaps, a demo pack to a trade show. The duty doesn't stay with the vendor.

So the current summary lives in my document control, revision-tagged, with the report identifier I copied at the start, filed next to the spec sheet it belongs to. Suppliers change portals. Suppliers get acquired. Suppliers rename part numbers. My file is the one link in the chain I control, and it's the thing that saves me when a question arrives eighteen months after the order closed.

The summary is not a certificate. It's a description of a battery that was tested on a particular day in a particular shape. The job is to make sure the one you're buying is still that battery — and to do it before you sign.