Four rings between the broadcast and the truck

The problem

The expensive part is not the invoice. It is the mark on the relationship.

The window between the customer signal and line-side delivery is ninety minutes to a few hours. Within it you must build the right variant, in the right order, defect-free, with material that may not have arrived. First ring, the sequence. The customer schedule defines which variant is due and in what order. Connect keeps that ring alive with what electronic data interchange does not carry: the supplier delay, the mix change announced by email, the issue the supervisor dictated while walking. If a component's coverage does not reach the window, you know hours ahead.

• A seat arriving out of sequence, with the wrong variant or non-conforming stops the assembly plant's line, and that is billed by the minute.
• The mark it leaves on the commercial relationship lasts years and is paid at the award of the next program.
• The window between the customer broadcast and line-side delivery is ninety minutes to a few hours: build the right variant, in the right order, defect-free, with material that may not have arrived.
• Each control exists in some form today, and the seat still escapes, because the failure is a concatenation: a delay nobody read, a fixture not confirmed, a wrinkle sampled out, a cart loaded in the wrong order. Each link passes its own check.

How it fits the IRIS system

Four rings that cross-check — sequence, preparation, build and loading.

Second ring, preparation. Edge confirms the line is in the right state for the variant due, fixture, kit, references and torque program, before start-up. If not, JIDOKA AI blocks it. Third ring, the build. The camera verifies trimming at cadence and the torque tool integration ties every safety joint curve to the serial number. A non-conforming tightening or a trim defect stops the unit before it enters sequence. Fourth ring, shipping. Before loading, each seat's serial number is matched against its position in the truck sequence. If they do not match, loading stops. And every unit travels with its dossier already built. That is the sentence that sums up the whole system: if the four rings do not agree, the unit does not move on. And it stops here, in your plant, where it costs minutes, not on the customer's line, where it costs the relationship. SMED AI closes the loop from the other side: using the changeover evidence, it shortens changeover time so the four rings are not paid for in cadence.

Ring one keeps the sequence alive with what EDI does not carry. Ring two confirms the line is in the state for the variant due. Ring three verifies trim and ties every safety tightening to the serial number. Ring four matches each seat against its position in the truck. If the four do not agree, the unit does not move on — and it stops in your plant, where it costs minutes.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · Sequence	Supply alert (3) + receiving (7) + supervisor headset (4)	Coverage of every component against the window, known hours ahead
2 · Preparation	Changeover validation (10) + shift start (1)	Fixture, kit, references and torque program confirmed before start-up
3 · Build	Trim defects (9) + torque traceability (8) + foam panel (2)	A non-conforming tightening or trim defect stops the unit before sequence
4 · Loading	Human sign-off (5) + the traceability master	Serial number matched to its truck position; if not, loading stops
Dossier	Lab tests (6) + evidence pack (11)	Every seat travels with its evidence already built
Cadence	SMED AI on the changeover history	The four rings are not paid for in cadence

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 6-12 months — it pays for itself with the first customer line stop avoided.
• A single sequence incident usually has a direct impact far above the cost of the full deployment, before counting the mark it leaves on the relationship.
• The discrepancy is caught in your plant with room to correct, and the evidence is available in minutes.
• And protection of the commercial relationship and the supplier rating, which is what the next program award is decided on.

Estimated payback 6-12 months · it pays for itself with the first line stop avoided The return is calculated against the first line stop avoided: a single incident of that kind usually has a direct impact far above the cost of the full deployment. Estimated payback of six to twelve months, an estimate to validate against your history of sequence incidents and the per-minute stop cost agreed with your customer.

And the fair question from the production manager

“Do we have to deploy all twelve cases to get this?” — no, and framing it that way is how you never start. The rings are built in layers and each case pays on its own. With the supply alert and the shift start you already have ring one, which is where the supply break lives; and none of it is a leap of faith, because every reading underneath is an anchored task where the best models drop below 1.5% error [1], with a person confirming at the gate.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.