UN 38.3 is the transport test series in the UN Manual of Tests and Criteria that a lithium ion, lithium metal or sodium ion cell or battery must be proven to meet before it can legally move by road, rail, sea or air. It is transport law, not vehicle law, and it applies to your first prototype shipment rather than your first production shipment. There are eight tests: altitude simulation, thermal cycling, vibration, shock, external short circuit, impact or crush, overcharge and forced discharge. The first five run in sequence on one sample, so damage accumulates. No government issues a UN 38.3 certificate. What exists is a ten-element test summary that manufacturers and distributors must make available on request. Development hardware can still move before testing is finished: Special Provision 310 in the UN Model Regulations excepts pre-production prototypes carried for testing, and production runs of not more than 100 cells or batteries, subject to strict packaging rules and, for air freight, a regulator's approval.
What UN 38.3 is, and when it bites
UN 38.3 is a set of eight transport tests in the UN Manual of Tests and Criteria. Lithium cells and batteries are Class 9 dangerous goods, and UN 38.3 is the gate that lets them travel. It reaches you through ADR and RID for road and rail, the IMDG Code for sea, ICAO and IATA for air, and 49 CFR 173.185 in the United States.
It is transport law, not vehicle law. R100 does the vehicle safety job, and the two share no test evidence. R100's vibration sweeps to 50 Hz in one axis, UN 38.3's to 200 Hz in three.
Here is what programmes get wrong. UN 38.3 is not a launch gate, it is a month-one problem. The first time a pack or module leaves your building it is a dangerous goods movement. Sending an A-sample to a customer, a test house or an integrator triggers it, and nothing about series production has happened yet. Automotive, bus, marine and off-highway programmes all hit this in month one.
A pack you cannot ship is a pack you cannot test.
Two more things. There is no such thing as a UN 38.3 certificate. A laboratory runs the tests and writes a report, and responsibility for the design sits with you. What exists is a ten-field test summary that every manufacturer and distributor must make available on request. If you are buying packs, ask for it by name, for the exact model, and check the Manual edition and the product description. A superseded edition invites a challenge, and a summary that does not match the box is as good as missing.
And buying tested cells does not give you a tested pack. A battery is tested in its own right.
The eight tests, and what each one stands in for
T.1 altitude is the unpressurised cargo hold of an aircraft, hunting for a seal that leaks or a cell that swells.
T.2 thermal is a container on a dock in July and a freight yard in January, swung between roughly 72 °C and minus 40 °C. It attacks joints, adhesives and anything whose stiffness moves with temperature.
T.3 vibration is the lorry, swept from 7 Hz to 200 Hz in three axes. Fixings, busbar terminations and cell retention fail here.
T.4 shock is being dropped by a forklift.
T.5 external short circuit is a conductive object across the terminals in a warehouse. Does your protection act first?
T.6 impact or crush is a cell penetrated or squashed.
T.7 overcharge is a charger that fails to stop.
T.8 forced discharge is a cell driven into reverse polarity by its neighbours.
Two structural details matter more than any parameter.
The first five run in sequence on the same sample. One unit walks through altitude, thermal, vibration, shock and short circuit, so a marginal design accumulates damage and fails late having been weakened early. By then you have spent the sample and the slot.
One pass criterion is easy to miss. Alongside no leakage, no venting, no rupture and no fire, the early tests require open circuit voltage after the test to be at least 90 per cent of the voltage before it. A pack can come off the rig looking perfect and still fail. T.7 also needs seven days of observation after a 24 hour test, so plan around the window.
Where the money actually goes
Everything you put into a UN 38.3 campaign, you destroy. Test articles do not come back. So the cost is not the lab invoice. It is two things: how likely you are to run the campaign twice, and what each unit cost to build. Both are design problems.
The first is confidence. A pack whose structure, interfaces, cooling design and protection architecture have already been engineered and already been through this test series enters expecting to pass. A clean-sheet pack enters hoping to. A repeat costs you the samples, the slot and the calendar.
The second is unit cost, and it is the one people miss. If every prototype is a one-off, with parts drawn and made from scratch for one programme, each test article is expensive, so a destructive campaign is expensive. If the pack is a configuration of an architecture that already exists, with parts, tooling and processes that already exist, prototype build cost falls and the same campaign gets materially cheaper.
The third effect is maturity. Because we run an architecture and platform technology approach, even our early prototypes can be at B+ sample maturity. A typical A sample is a first go at packaging, thermal path and structure, built to learn from rather than certify on, so testing on it usually has to be repeated later. A configuration of an already-engineered architecture starts closer to production intent, so what goes on the rig is representative from an early build and testing moves forward in the programme.
Shipping prototypes before you have a test report
The regulations anticipate exactly this problem, and almost nobody writes about it. Special Provision 310 in the UN Model Regulations, carried into ADR, RID and the IMDG Code, disapplies the UN 38.3 testing requirements for two things: production runs of not more than 100 cells or batteries, and pre-production prototypes transported for testing. Both are conditional on packing instruction P910, or its large-packaging equivalent LP905 for heavy items like traction packs.
The two limbs differ. The prototype limb is purpose-limited: the text says transported for testing. Moving a prototype so someone can test it is inside the provision. Moving one so a customer can drive a demonstrator is not, or at least treat it as outside until a dangerous goods advisor says otherwise. The 100-unit limb has a volume limit, not a purpose limit, which makes it more useful for anyone building in the hundreds.
Neither limb is a free pass on packaging. The packing instructions are the real content of the exception: individual inner packagings, non-combustible and non-conductive cushioning, movement prevented inside the package, outer packagings to Packing Group I performance.
Air freight is where it gets tight. Untested prototype and low production run cells and batteries are barred from passenger aircraft, and allowed on cargo aircraft only with prior approval from the relevant authorities, under Special Provision A88. In the UK the Civil Aviation Authority runs an application process for this. Build the lead time in, because it is a regulator's queue, not a forwarder's.
Treat this as a plan, not a loophole. It moves development hardware, not a programme.
Change the design and you start again
The Manual defines a type as a particular electrochemical system and physical design. Change the electrochemistry, or change the physical design in a way that could cause the failure of any test, and you have a new type that needs testing.
Until a test house says otherwise, treat these as triggers: a different cell, a different chemistry, a different cell supplier, a change to module or pack structure, a change to the electrical architecture, a change to the protection devices.
R100 bites the same way. Every modification to an approved battery type must be notified to the Type Approval Authority that granted it, which can revise the approval, extend it, or require a new one. An extension can mean going back to the test house.
So a late design change hits you twice, and both hits need hardware you build and destroy again. Getting it right first time beats fixing it later.
How we handle it
We do R100 approval on all battery systems as a standard part of our development programmes. UN 38.3 is done the same way: standard programme scope, not an option, not a later phase, not something you commission yourself.
Three things follow. You start with confidence of passing, because the pack is a configuration of an architecture already engineered and already tested. You spend less on the hardware you destroy, because the prototypes are cheaper to build. And you deal with one partner for design, build, test and approval, instead of coordinating a design house, a fabricator, a test house and a dangerous goods advisor.
The part most people find out too late is how A and B samples move before any test report exists. If that is in front of you now, get in touch and we will map the transport route alongside the design.
