The plant takes two years. Judgement takes ten.

Here is an uncomfortable observation about this week's manufacturing news: every headline is a cheque. A $112 million expansion at an EV plant in Kentucky. A $770,000 grant to a southern Oregon college to train manufacturers. A Rocky Mountain region asking, in print, who will build its future. And an Indian EV maker with a four-month waiting list, while another watches its recalls collide with its global push. Capital commits in a board meeting, sometimes in a single afternoon. Capability does not. Nowhere in these announcements is there a line item for the person who will stand in front of a suspect lot at 05:40 on a Thursday, a truck at the dock and a shift pressing behind them, and decide whether it ships.

What the training grant actually buys

Grants like that are good news. Be precise about what they buy, though: courses, contact hours, certificates – laminated, often, which is a nice touch. What they cannot buy is the disposition call. The feel for a dimension that is inside tolerance but drifting, on a part that changed supplier last month, on a Friday afternoon. And the spine to treat it as a problem before the data proves it is one.

I learned this the expensive way, building a greenfield QA/QC department for a 900-person plant from zero: no inspectors, no engineers, no inherited habits – which sounds like a problem and is actually a luxury. The résumés arrived fine. Certified, schooled, fluent in the core tools. Judgement did not arrive with them. So in interviews we stopped asking about tools, put a real deviating part on the table and asked one question: "Does this ship?" Ninety seconds separated the practitioners from the students. Some of the finest tool knowledge I have ever seen sat inside people who could not make that call.

The grant buys the course, the course produces the résumé, and the résumé walks onto the floor to meet its first nonconformance that looks like three others but isn't. That gap – between knowing PFMEA and exercising judgement – is what nobody is budgeting for.

Growing judgement at line rate

You cannot lecture judgement into people. You apprentice it, one real nonconformance at a time. We ran a tiered structure on purpose. New inspectors verified: measured, documented, compared to spec. Disposition authority sat a tier above and was earned, not conferred – the criterion was blunt, roughly twenty live QRQC cases closed under supervision, at least two of them escalations. The third tier, the nerve to wake a customer quality engineer at 02:00, came last. Fewer people carried it than the org chart assumed.

QRQC was the vehicle, not a ritual. Every live defect was a teaching case: frame the problem in fifteen minutes, go to the part, see the failure mode with your own eyes, propose containment within the hour. Rotation was deliberate. Everyone did time in the containment pen, everyone chased a supplier escape, everyone sat in on a customer complaint call. You cannot simulate the texture of the moment a customer asks how the part left your building and the answer has to be true. You can only be in the room four or five times before it is your question to answer.

Capital can be committed in an afternoon. Judgement has a ten-year lead time, and nobody stocks it.

Controls that assume the gap

Until competence compounds, the system has to compensate – and it can, if you design it honestly. Nothing left quarantine without a disposition-tier signature. A critical characteristic drifting toward its control limit across three consecutive lots pulled an engineer inside thirty minutes; the trigger was binary, so there was nothing to argue about. And the newest hire's menu had exactly three options: verify, hold, escalate. Design it so the worst thing a beginner can produce is a held lot – an hour of delay – never an escape.

The trade-off is real and should be priced like adults. Slower release, inspection overtime: we ran 100% inspection on safety characteristics for the first six months of ramp, and I would do it again. Call it €40,000 a month in hours and delay for a mid-size line. One escaped lot that stops a customer line burns through that before lunch – sorting, containment, expedited freight, goodwill – and that is the cheap version, the one where nobody gets hurt. Later, leading quality initiatives across a 2,000-person, multi-site workforce, the pattern held: the sites that front-loaded the controls ramped fastest, because they spent their delays early instead of their credibility late.

So read the expansion announcements with respect, and with arithmetic. Treat quality capability as a production system in its own right – its own lead time, its own ramp curve, its own capacity plan – not a hiring line on an org chart that fills itself because a course exists. The sequencing is unforgiving. The building gets enclosed, the robots get tuned, the launch date gets met. The plant takes two years. Judgement takes ten. Start growing it now, or accept that you have designed escapes into your own ramp – and you will find out at 05:40 on some Thursday, with the truck at the dock, that nobody in the building has the nerve to say stop.