Three EVs a day. Validation doesn't scale.

A plant has two kinds of bad morning. The first is an escape: one defective part in a return bin, a containment tag, a long afternoon. The second is a batch: the shipping lane goes quiet, nobody can say which pallets are good, and the honest answer – none of them, and we don't know since when – arrives by phone. I have lived both. They are not the same species of problem, and treating them as if they were is how plants lose quarters.

So when CATL – the world's largest battery maker – says the words "batch failures" in public, I don't hear a warning. I hear a forecast. And when it lands in the same news cycle as China launching nearly three new EVs a day, these are not two stories. They are one story about an industry consuming validation capacity faster than it builds it.

An escape missed a control – a batch never had one

An escape is a control that blinked. Something existed – a poka-yoke, a torque monitor, an end-of-line test – and one part got past it. Painful, traceable, fixable. A batch nonconformance is different in kind: the control that should have validated that process configuration never existed, or was never exercised. If four thousand cells from one shift share a defect, the useful question is not how final inspection missed them. It is which gate was supposed to prove that process change, and why it never closed.

Sampling hides this. AQL plans assume defects are independent events scattered across a stable process, so thirty clean parts at start-up tell you something about part thirty-one. Batch failures are correlated. The first two hours run fine, the drift starts when the oven reaches steady state or the second lot of material is opened, and your sample – taken at commissioning, when the line was on its best behaviour – says nothing. Sampling is built for the day a part misbehaves. It is blind to the day the process decides to.

An escape is a control that blinked. A batch failure is a control that never existed.

The economics are asymmetric. A single escape costs sorting, premium freight and an 8D – call it €20,000 to €80,000 and a bruised quarterly review. A batch nonconformance starts at six figures in scrap and reaches seven once you add supplier containment, line-down penalties and the customer's sorting crew. I have watched one batch event erase a year of a plant's cost improvement plan. That asymmetry alone makes batch prevention its own discipline.

Three launches a day eats the validation queue

Every new EV is not one product; it is a stack of launches. The OEM launches a vehicle, the tier-one launches a pack, the cell maker launches a chemistry revision, and every layer draws on the same resources: APQP hours, PFMEA workshops, PPAP packages, durability and thermal-runaway cycles, EMC, firmware. Multiply that by nearly three vehicles a day in China, add the "minor" changes riding through existing part numbers, and validation becomes a queue like any other. When the arrival rate exceeds capacity, work does not wait politely. It compresses.

The compression has a signature. The FMEA is last year's model with the dates changed – I can spot one in four minutes, because its top-ten failure modes still list a component the programme deleted a year ago. The durability test that ran 1,000 cycles is "correlated" to 300 plus simulation. The supplier's process change ships before requalification because SOP moved up a fortnight. Nobody announces that validation was skipped. The memo says "risk accepted, monitoring in place".

Budget validation like press hours, not paperwork

I learned this at SNOP, building the QA/QC department of a greenfield plant that grew past 900 people, with launches arriving in seasonal waves. The instinct everywhere is to treat validation as paperwork with a deadline. We did the opposite and budgeted validation the way production budgets press hours: engineer-hours and rig-hours per variant, a published capacity, queue time reported weekly beside output. Once the queue was visible, the conversation stopped being "push it through" and became structural. Move the gate, add rig capacity, or move the launch with the customer – in writing.

Nonconformances got the same treatment. Every batch signal opened a QRQC case inside 24 hours, with an A3 on the system condition behind it. That discipline, not any inspection we added, is where the 70% reduction in defect costs came from. At WITTE Automotive before that, A3 countermeasures and Q-Wall gates on change management did the same job: no change reached production without requalification evidence attached. I have led quality initiatives across multi-site IATF 16949 operations with more than 2,000 people, and held a quarter to zero critical customer escalations. Not heroics. Batch thinking, applied before the dock.

None of this is exotic under IATF 16949. Change management already obliges you to define requalification triggers – material, tooling, process parameters, supplier, site. The failure mode is not the standard. It is enforcing it on a Friday afternoon three days before SOP, and accepting a supplier's notification of change after the parts have shipped instead of before. A change you first hear about from a bad batch arrived three gates too late.

The warranty ledger decides

CATL's statement is the leading indicator; warranty data is the lagging one, reporting 18 to 30 months after SOP. The batch failures that will shape the EV shakeout are being signed off right now – at three launches a day, in FMEA workshops nobody has time to run properly. The market will not be settled by whoever launches the most models. It will be settled by warranty accruals. The survivors will be the plants whose validation capacity grew in step with their launch cadence. Planned and gated like capacity, batch nonconformances stay events. Left as paperwork, they become the sector's operating rhythm – and validation, at this cadence, is capacity – which is, I suspect, exactly what CATL was telling us.