Four verification rings, from the tanker to the container

The problem

Reputation, certification and export plan: three assets one incident can touch at once.

Reputation, certification and export plan are three separate assets that a single incident can touch at once. A recall of exported product combines direct collection and destruction costs, commercial channel blockage, re-certification costs and a reputational cost in the destination market that is multi-year and hard to bound. And there is a second problem, less dramatic and just as real: a plan to double export volume built on manual processes requires doubling the administrative headcount that sustains them. Growth eats its own margin. The eleven preceding cases each solve their own piece. What is missing is the coordination: making it impossible for a lot to leave with any piece unverified.

• A recall of exported product combines direct collection and destruction costs, commercial channel blockage, re-certification costs and a reputational cost in the destination market that is multi-year and hard to bound.
• The worst day is not a high-scrap day. It is the day a container is held at the border, or a lot is withdrawn abroad, or a re-inspection raises a non-conformity that puts the certification opening that market in question.
• And the less dramatic problem: a plan to double export volume built on manual processes requires doubling the administrative headcount that sustains them. Growth eats its own margin.
• The eleven preceding cases each solve their own piece. What is missing is the coordination: making it impossible for a lot to leave with any piece unverified.

How it fits the IRIS system

Four rings that must all be green — it is all of the above, orchestrated.

Connect + Edge + Agents + legacy integration + JIDOKA AI + SMED AI, coordinated against a single objective, in four rings:

The flagship adds no new technology. No single ring solves the problem on its own: each already exists today in some form and the lot still escapes. The strength is in the sequence — the lot cannot reach ring four with ring two unverified — and in the fact that a missing piece is flagged and resolved inside the site, where it costs a hundred times less than at customs.

• Ring 1 — Intake (cases 5 and 7): no tanker is discharged and no input pallet enters without human validation over consolidated data. The batch origin is born traced, not typed.
• Ring 2 — Process (cases 1, 2, 6 and 10): the startup sheet, the sterilizer and evaporator curves, the lab analysis and the visual changeover validation are tied to the batch as they happen. JIDOKA AI blocks startup if the changeover is not verified.
• Ring 3 — Outlet (cases 8 and 9): the coder prints what the ERP says and confirms what it printed; the Edge camera inspects one hundred percent of seals and coding. No badly sealed or badly coded pack enters the pallet.
• Ring 4 — Dispatch (case 11): before the pallet leaves the warehouse for the container, Agents checks that the lot has its complete dossier for the destination market and issues it. If a piece is missing, the lot is not dispatched: it is flagged and resolved inside the site, which is where it costs a hundred times less.

SMED AI cuts across all four rings, accelerating changeovers between batches, because the argument cannot be "safer but slower".

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 · Intake	Tanker release (5) + goods receipt (7)	No tanker discharged and no input pallet in without a human signature
2 · Process	Startup (1) + evaporator (2) + lab (6) + changeover (10)	Sheet, curves, analyses and changeover tied to the batch as they happen
3 · Outlet	Coder (8) + seal inspection (9)	No badly coded or badly sealed pack enters the pallet
4 · Dispatch	Evidence pack (11)	No lot leaves without its complete dossier for the destination market
JIDOKA AI	Across rings 2 and 3	Blocks startup if the changeover is not verified
SMED AI	Across all four	Faster changeovers, so it is never safer but slower

from detection that happens outside — on the shelf, at customs, in the audit — to detection that happens inside, at the relevant ring, with the product still at home. From an export doubling plan that requires doubling administration, to one that scales without it.

Impact estimate

Estimated impact — to validate with quality and general management.

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 8-15 months in the conservative scenario, counting no avoided incident at all.
• Counting the first one — a held container, a withdrawn lot, a non-conformity at re-inspection — effectively immediate.
• Detection moves from outside — the shelf, customs, the audit — to inside, at the relevant ring, with the product still at home.
• And an export doubling plan that scales without doubling administration, because the evidence is generated as a by-product of producing.

estimated payback 8 to 15 months in the conservative scenario, counting no avoided incident at all. Counting the first one, effectively immediate. *Estimate to validate with quality and general management.*

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 would be the way never to start. The rings are built in layers and each case pays on its own from the first month. With tanker release and goods receipt you already have ring one; add the startup sheet and the lab sheet and half of ring two is closed, which is where most of the missing traceability sits. Where a ring depends on a model reading a document or an image, that is an anchored task with the best models below 1.5% error [1], and each ring ends in a human signature.

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