The state audits the line. You audit the binder.

When four government departments in China jointly launch a special campaign on the production consistency of road motor vehicles, most executives reach for a binder. Wrong reflex. The inspectors do not read your quality manual. They pull a serial vehicle – sometimes one still on the line, sometimes one already sold – and diff it against the homologation file: material against material, calibration against calibration, label against label, software version against the version the state actually approved. Not against your intent. Not against the current drawing. Against the frozen record from years ago. Two decades of automotive and aerospace quality work have taught me how that comparison usually ends. Most audit trails cannot survive it.

What production consistency actually demands

Conformity of production asks a narrower, harsher question than quality. Quality asks whether the part meets the requirement. Conformity asks whether it is the product the state approved. The approved product, not a good product. A part can be cheaper, more durable, even safer than the approved version and still be nonconforming, because the homologation file names a specific resin grade, a specific calibration, sometimes a specific supplier – and the unit in the inspector's hands is running something else.

Engineers are trained to improve things, which is precisely the problem. Every improvement is a deviation until it is re-approved. IATF 16949 knows this: clause 8.5.6 forces control of change and re-verification. PPAP exists to decide when a change forces re-submission. VDA 6.3 auditors probe change-point management for the same reason. The framework is not missing. What is missing in most plants is the habit of treating the approval file as a live baseline instead of a trophy on the wall.

Type approval is a photograph of your best prototype. Conformity of production is the unedited footage of every shift since.

Where consistency quietly dies

Consistency never dies in a meeting. It dies in a hundred small, confident decisions on the shop floor. Across two decades of QRQC sessions in automotive and aerospace, the root cause was an unvalidated running change more often than any operator error. A refurbished forming tool comes back with a re-cut cavity. The parts still sit inside tolerance, so nobody requalifies, and the dimensional report in the approval file now describes a process that no longer exists. Or a sub-tier swaps a steel grade for an equivalent one, certificate in hand, and tells nobody upstream. The part passes every incoming check. Containment later runs past €150,000. The material approval in the dossier still names the old grade.

The shortest A3 I ever signed ended at a five-why that read like an epitaph: the change was never classified as a change, so it was never a change. The part worked. The part always works. "The part works" is not a defence, because it was never the question.

Leading quality initiatives across a 2,000-plus workforce on multiple sites, I watched two plants drift from one standard in slow motion – a local substitution here, a locally tightened weld parameter there. Same part number. Two different products within two years. Pull a serial from each site and only one conforms. The certificate does not know the difference. The inspector will.

Building a line the state can audit

At SNOP I got a rare luxury: an empty building, a 900-plus-employee greenfield plant, and a mandate to write the quality system – change control, traceability, configuration baselines – before the first serial part existed. That sequencing is everything. You cannot retrofit an audit trail onto a decade of tribal knowledge; the record must be created at the moment of change, not reconstructed after the regulator calls.

Three mechanics carry the weight:

• Change-point management that classifies at birth. Every change is stamped approval-relevant, safety-relevant or cosmetic the moment it is proposed, and the classification – not someone's optimism – decides whether the path is a declaration, a PPAP re-submission or full re-validation.
• A configuration baseline per lot. Any serial unit must be reconstructible: which tool, which cavity, which material lot, which software release. In days, not months.
• Risk-triggered re-validation, not calendar-based. An annual conformity audit samples whatever runs that week. The event that demands re-validation is a change to the approval scope, not a date on a wall planner.

The campaign will end in a few months; the discipline should not. Plants that treat conformity as daily practice barely notice inspectors, because their normal week is the audit. Plants that prep for the event will pass this campaign and fail the next unannounced pull, having fixed the binder instead of the line. When enforcement shifts from certifying documents to diffing serials, your change-management history stops being paperwork and becomes your only defence. The state audits the line. Audit it first.