The Eight UN 38.3 Tests Behind Your Robot Pack's Test Summary

A UN 38.3 test summary is a verification document, not a formality, and it gets read in three passes: the test list, to confirm the page covers the tests that apply to a battery pack rather than to bare cells; the description block, to confirm the tested configuration is the configuration on your purchase order; and the date and report trail, to confirm nothing has changed since the test that would leave the document describing a pack you no longer buy. If any of those three fails, the results column is decoration.

I keep two things open whenever a summary lands in my inbox: the PDF, and the pack drawing with the current revision block. The summary arrives as a two- or three-page scan, sometimes on a lab letterhead, sometimes on the manufacturer's letterhead with the lab named in a footer. I don't start with the results. I scroll to the description, because a results column only means something if the rows describe the object that's going to sit in the crate.

Why you're handed the summary and not the report

UN 38.3 is the test regime that lithium cells and batteries pass through before they can move as dangerous goods. The full test report behind it is a thick document: chamber logs, photographs of the unit before and after, the serial number of the sample, the raw voltage and temperature traces. You will almost never see it before a purchase order. What travels is the summary, a one- or two-page condensation built to be handed to a freight forwarder, a carrier, or a downstream customer's EHS person without disclosing anything proprietary about how the pack is built.

That's the design intent, and it's a reasonable one. It also means the summary is the only window you get, so you have to read it as evidence rather than as reassurance. One habit pays for itself immediately: write the test report number into the PO file even when you don't receive the report itself. The number is what lets you go back later, and it's what the forwarder will ask you to read aloud.

The eight tests, and the two that don't show up on a pack sheet

UN 38.3 runs eight numbered tests. T.1 is altitude simulation, a low-pressure soak. T.2 is the thermal test, cycling between temperature extremes. T.3 is vibration. T.4 is shock. T.5 is external short circuit. T.6 is impact and crush. T.7 is overcharge. T.8 is forced discharge.

When I open the test table on a pack summary, I expect six rows: T.1 through T.5 and T.7. T.6 and T.8 are cell-level tests, which is why they belong on the cell summary and not on the pack sheet. A pack summary with only six lines is not a pack summary missing two lines. This trips up more buyers than anything else on the page, because "eight tests" is the number that sticks in everyone's head.

Which means the other document matters. If the summary references a separate cell-level test report by number, that's a good sign — someone assembled the paperwork deliberately. A pack summary with all eight rows is worth a question: sometimes the lab really did run impact and forced discharge at the pack level, and sometimes a cell sheet got merged into the pack sheet so the page would look complete. Ask which document the pack results are actually recorded in. Any row that reads "Not applicable" or "N/A" deserves the same question.

On the results themselves, I'm looking for a clean pass on every row, with the acceptance criteria behind it: no mass loss beyond the limit, no leakage, no venting, no disassembly, no rupture, no fire, and open-circuit voltage holding above ninety percent of the pre-test value. Those are the things the test is actually watching for. A table of green checkmarks with no criteria stated anywhere on the page is a summary someone typed, not one someone wrote.

The description block is where most summaries are won or lost

This is the part I read three times. The description has to identify the manufacturer and address, the cell, the pack's mass and watt-hour rating, and a physical description. Then I lay it against the drawing, line by line.

The cell, by full part number

"21700" is a size. It is not a cell. If the summary says the pack uses a 21700 cell and my drawing says a specific vendor part number with a suffix, those are not the same statement. Cell capacity bins, tab configuration, and internal protection vary across the same format, and the thermal and vibration response varies with them. I want the exact part number on both documents, matching character for character.

Watt-hours and mass, checked against my own math

Watt-hours are nominal voltage times rated capacity. If the drawing says 48 volts nominal and 20 amp-hours, that's 960 watt-hours, and a summary claiming 1,050 is describing a different build — possibly a max-voltage-times-capacity figure, possibly a different cell. Mass is the other fast check. A pack gets weighed for shipping anyway, so a mass difference of more than a few percent between the summary and the drawing isn't measurement noise. Something inside the enclosure changed.

Cell count and arrangement

A 7S4P pack is not a 7S3P pack with a footnote. Energy, mass, current per cell, and the vibration and shock response all move. If the summary doesn't state the configuration, that's a gap on the page, not a detail you can assume.

BMS and protection devices

Many summaries never mention the BMS. That's a problem on a pack, because the BMS is usually what terminates an overcharge event — the protection threshold opens the FETs before the pack reaches a state the test is designed to catch. So when a summary shows a clean T.7 pass and says nothing about the BMS, I ask which BMS was in the tested unit and whether it's the one in my build. Same logic for a fuse, PTC, CID, thermal cutoff, or vent. These are not accessories. They are the parts that made the test pass.

Enclosure

Material, wall thickness, venting, and the thermal path through the housing all feed into T.2 and T.6. If the tested enclosure was machined aluminum and mine is a stamped steel box with a different vent, the document and the product have drifted apart.

How to spot a summary written around a representative pack

Every supplier has one hero unit that got tested, and a production line that has been quietly improving since. Most of the time nothing bad comes of it. Sometimes the page is describing a pack that was never built twice.

The tells are usually in the wording. "Representative of," "family," "series," "similar construction," "type" — those phrases mean the described unit stands in for a range, and you need to know how wide the range is. A model number that's a product family name rather than the part number on your PO is the same signal. So is a photograph showing a connector, a mounting foot, or a label that doesn't match your drawing, because photos in these documents are almost never staged.

The fastest single check is the calendar. Put the test date next to the drawing revision date. If the drawing was revised after the test and the revision wasn't cosmetic — a label change, a silkscreen tweak — find out what moved. It is a thirty-second comparison that has saved me more time than any other habit on this list.

Who issues it, and what has to be on the page

The summary is normally put out by the manufacturer of the battery, drawn from a test report produced by a testing laboratory. Both names should appear. So should the lab's address, the report number, the test date at least to the month, the description of the unit, and the list of tests with results. If the summary is issued by a trading company or an assembler who bought cells and packaged them, I want to know which entity is the manufacturer of record, because that is the name that will end up on the shipping papers and the name a carrier will look for.

Contact information is a small thing that says a lot. A street address and a company domain carry differently than a personal email account, especially on a document that may be read by someone who has never heard of your supplier and has no way to trace them.

Changes that send you back for a test

I don't decide whether a change requires retesting. The lab does, and I say so out loud, because that keeps the conversation about the build instead of about trust. My job is to notice that a change happened and to hold the shipment until someone with the equipment and the report in front of them confirms the existing document still covers what's on the pallet.

The changes that reliably start that conversation:

A different cell — different vendor, different part number, different capacity, or a capacity bin change within the same part number.

A different cell count or arrangement.

A new BMS, a BMS firmware change that moves the overcharge cutoff or a protection threshold, or a change in which conditions trigger the cutoff.

A protection device added, removed, or swapped for a different rating.

An enclosure change that alters the thermal path or the crush behavior.

A change in how the cells are joined — spot weld to laser weld, wire bond to busbar. The interconnect is part of the mechanical structure that survived vibration and shock.

A change of assembly site, when the process changed along with the address.

A change in mass or watt-hour rating beyond the tested unit. This is the one that hides best, because it arrives as a quiet upgrade: same part number, slightly higher capacity, and a test summary that no longer describes the product. If the original cell goes end-of-life and a substitute is qualified, that substitution is a retest trigger even if the part number on the PO never changes. That's a continuity risk that surfaces two years into a fleet, not at the PO.

The same page is a shipping document

It's easy to file the summary under procurement and forget what it's for. This is the document that supports the pack moving as UN3480 when it ships by itself, or as UN3481 when it ships with or inside equipment. It goes with the shipment to the forwarder. It stays in your file for the life of the fleet, because three years from now an EHS manager, a carrier, or a customer who is shipping the robot onward will ask for it, and "we'll get it from the supplier" is not an answer you want to be giving under time pressure.

That's why configuration drift matters more than it looks. A summary that stopped matching the product doesn't just weaken a purchase file. It weakens the document that every future shipment depends on.

So the checklist I keep is short and it lives in the PO folder: six rows on the pack sheet, both summaries present, cell part number matched, watt-hours checked against my own multiplication, mass within a few percent, BMS and protection devices named, report number recorded, test date compared to drawing revision. When a supplier pushes back and says the summary is standard, I tell them I'm not questioning the test — I'm matching the paperwork to the bill of materials. That framing has never cost me a supplier, and it is the cheapest check in the entire procurement process.