The battery pack warranty gap that appears in year three

The gap opens the moment a capacity fade guarantee is written as a sentence instead of a test. "Eighty percent of rated capacity at three years" promises nothing until six things are fixed in writing: the cycle count, the charge and discharge rate, the depth of discharge, the ambient temperature, the measurement method, and the capacity threshold that counts as end of life. Leave any one of those blank and the disagreement does not get settled at the bench where it could be — it arrives in year three, as an argument about whose test was the right one.

The pack on my bench came back in a plastic tote with an RMA slip taped to the lid. Low runtime, no fault codes, BMS log clean, two and a half years on an AMR fleet. The route plan said one charge cycle per shift. The charge log said otherwise.

I clamp it to the cycler, tape a thermocouple to the case, hold the room at 25 C, and discharge at 0.2C. It comes back at 82 percent of nameplate. The supplier's warranty language says 80 percent at three years. Their test, their number, their win.

Then I run it the way the robot actually runs it. Chamber at 40 C, discharge at 1C, full depth because the fleet charges to 100 percent and runs to cutoff. Same pack, same afternoon, 74 percent.

Nothing dishonest happened in that room. Two tests, two answers, and only one of them is in the document. That is the whole of the year-three gap — it is not a defect, it is unassigned test conditions that both sides leave alone because the language sounds generous and nobody wants to slow the PO down.

I have watched that same conversation happen on four programs now, and it always rhymes. The buyer thinks they bought a fade guarantee. The supplier thinks they sold a warranty against a standard test. The gap is the space between those two sentences, and it is almost always wide enough to drive a fleet through.

Cycles per day, not years

Warranties are written in years because procurement calendars run in years. The wear mechanism does not.

Three years is roughly 1,100 days. At one cycle a day, that is 1,100 cycles — often under what the cell will do before it hits 80 percent, which means the term is doing less work than the buyer assumes and calendar aging might be the real limit. At two and a half cycles a day, which is what opportunity charging between missions produces, you hit 1,100 cycles in about fourteen months and then you are carrying twenty-two months of paper coverage on a pack that already passed the only threshold that mattered.

The guarantee has to state both a cycle count and a term, and the cycle count has to come from your duty cycle, not from the supplier's standard test protocol. I build that number from the fleet simulation: missions per shift, energy per mission, usable window, charging opportunity. Then I put it next to whatever the supplier will actually warrant. If mine is higher, I am self-insuring the tail and I want that stated plainly on the internal approval rather than discovered later.

One detail that bites: partial cycles. Opportunity charging produces 0.3-cycle events, and a BMS may count them by coulomb throughput or by discrete cycle events. Which counting method decides "cycles consumed" is itself a warranty term. I have seen a fleet burn through a cycle allowance in eighteen months on one counting convention and still be at 60 percent of it on another.

Temperature and depth of discharge travel together

A fade guarantee that names a temperature but not a depth of discharge, or the reverse, is not a guarantee. They interact, and they interact in the direction that hurts.

As a rule of thumb I use for sizing, every 10 C of sustained cell temperature roughly doubles the fade rate, and pushing depth of discharge from 80 percent to 100 percent costs more capacity per cycle than the ratio suggests. Combine a hot aisle with a deep cycle and the fade curve steepens faster than either factor alone predicts. That is not a laboratory curiosity — it is what an AGV in a warehouse mezzanine in August does every day.

So the useful thing to ask for is not a single number. Ask for a fade table: at least two temperatures and two depths of discharge, so you can interpolate your own duty cycle honestly. Suppliers who have run the cycling can produce that in a week. Suppliers who have not will send a curve with no axis labels and a lot of confidence, and that tells you something about the audit you are about to run.

What the exclusion page assumes

Read the exclusion list against your real duty cycle while you still have the site data open. Over-charge, unauthorized chargers, operation outside the specified temperature range, water ingress, physical damage. Every one of those is reasonable on its face and dangerous in combination with a datasheet envelope of 0 to 45 C when the mezzanine hits 47 C in July.

If the BMS log is the arbiter of a claim, make sure you can pull it and that it retains history for the whole term. A BMS that keeps rolling thirty-day data cannot defend a three-year argument, and I have seen exactly that turn a legitimate claim into a goodwill conversation.

The threshold is meaningless without a measurement method

Write the failure criterion as capacity below a stated share of a stated baseline, and then define the baseline. Nameplate is not a baseline. A pack that ships at 103 percent of nominal has handed itself three points of free fade before the threshold ever bites, and the buyer paid for that headroom.

On our side of the table, the practice that has held up is to measure at receiving. Full capacity test at a defined rate and temperature, log it, and write the threshold as a percentage of that measured number. Now both sides are arguing about the same starting point.

Then fix the test itself. A 0.2C capacity check flatters any pack. If the robot discharges at 1C, the warranty test should be run at something close to the duty cycle, or at both rates so the divergence is visible. Rest period before the measurement, the temperature at the time of test, the equipment and who owns its calibration, whether the supplier accepts your lab's data or insists on their own — all of it belongs in the document, because in year three the only question anyone asks is who runs the test and on what.

Capacity is not the only way a pack fails

I have pulled packs that still tested at 86 percent of nameplate and were unusable in the field. Voltage sag under a hill start, drive faults on acceleration, a robot that browns out and reboots mid-mission. That is internal resistance, and a capacity threshold alone will not catch it.

Add a resistance limit to the failure criterion: a DCIR pulse at a stated state of charge, a stated temperature, a stated pulse duration, with a defined rise over the receiving baseline that counts as failed. It is one more line in the spec and it is the line that actually matches the complaint the technician writes on the ticket.

Where the clock starts, and who pays for the swap

Shipment date and installation date can be months apart on a fleet rollout. Packs arrive in batches, sit in a warehouse, get installed as robots get commissioned. If the term starts at shipment, every month of integration is a month of coverage you paid for and never used.

Ask for installation date, or ship date with a defined allowance and a defined storage condition — state of charge and temperature — so the supplier cannot later argue that your warehouse caused the problem. And ask what happens to the replacement: does its warranty restart, or does it inherit the remainder of the original term? Inheriting is common, and it means a pack replaced at month 34 is covered for two months.

Prorated versus full replacement changes the shape of the year-three exposure more than the term does. A prorated schedule pays out a shrinking fraction; full replacement puts a new pack on the dock. Neither is wrong, but the buyer needs to know which one they hold before the first failure, not after.

Freight and labor are the part everyone forgets. The pack is rarely the expensive part of a field swap. Someone has to decommission the robot, crate a heavy pack, ship it, and reintegrate. If the supplier ships a drop-in replacement to the site before the failed unit comes back, the fleet stays up. If the unit has to return first, the robot is down for the transit time twice, and no capacity threshold compensates for that. Get the advance-exchange question answered in writing alongside the fade numbers.

Regional stock and the response clock

A service level that promises a response time but no regional replacement stock is a promise to send an email. Ask where the spares sit, how many, and what the swap clock is measured from — fault call, RMA issuance, or receipt of the failed unit. Those three starting points produce very different downtime.

What a pilot can and cannot prove

A six-month pilot at two cycles a day is roughly 360 cycles. That is enough to catch a bad batch, a bad weld, a bad cell supplier. It is not enough to validate a three-year fade curve, and anyone telling you otherwise is selling.

What the pilot actually validates is the measurement discipline. Baseline capacity at receiving, a defined test at the end, DCIR pulse at both ends, cell-level temperature logging rather than pack-exterior logging, and a BMS history export that you have actually parsed yourself. If the pilot can produce those four artifacts, the fade guarantee has something to rest on and the year-three conversation has a shared vocabulary. If it cannot, you have learned something more valuable than a fade number.

The gap closes in the RFQ, before anyone negotiates price. It closes when the warranty reads like a test procedure — condition, method, threshold, arbiter, remedy — instead of a sentence about confidence in the product. That costs nothing at the buying stage and it is the only thing standing between you and a tote on a bench with an RMA slip taped to the lid.