Four coordinated verification rings, across both plants

The problem

A migration hold is never one failure: it is four records that never met.

A single incident destroys the packer client's trust, forces a precautionary hold across both plants, and puts the BRC certification at risk for the whole portfolio.

• The biggest risk for a food-contact packaging maker with two plants is a hold triggered by poorly traced varnish migration or adhesion.
• The work order, the curing curve, the lab result and the code on the can each live in a different place, and nothing checks them against each other. Each looks correct on its own, which is exactly why the mismatch is not seen until the packer client sees it.
• A single incident destroys the packer client's trust and forces a precautionary hold across both plants, because nobody can bound the affected runs.
• And it puts the BRCGS certification at risk for the whole portfolio, not just for the run in question.

How it fits the IRIS system

Four rings across both plants — Connect, Edge and Agents orchestrated with JIDOKA AI.

Ring 1 (ERP): batch and varnish spec per work order. Ring 2 (curing oven): digitized curve linked to the run. Ring 3 (lab): migration result automatically cross-checked. Ring 4 (coder/Edge): what's printed is compared against the active work order. If anything doesn't match, JIDOKA AI stops the line; SMED AI speeds up the switch between chemical and food-contact references.

No ring is new: each one is built from the previous cases. The protection comes from the cross-check, because a run that passes each control separately can still fail when the work order, the curve, the lab and the code disagree.

See the full IRIS architecture →

Before and after

The four rings, the cases behind them and what each one closes

Ring	Which cases it draws on	What it closes
1 · ERP	Run startup (1) + coil receiving (7) + validation (5)	Batch, coil and varnish spec per work order
2 · Curing oven	Oven panel (2)	Digitized curve linked to the run
3 · Laboratory	Migration spreadsheet (6)	Migration result cross-checked automatically
4 · Coder and Edge	Coder (8) + print vision (9)	What is printed, compared with the active work order
Stop and switch	Changeover cleaning (10)	JIDOKA AI stops on a mismatch; SMED AI speeds up chemical-to-food switches
Evidence and context	Evidence pack (11) + alerts (3) + earpiece (4)	One dossier per pallet, across both plants

1-3 incidents/year with partial hold risk → zero. Discrepancy detection on the line, not at the packer client.

Impact estimate

Impact estimate — to be validated 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 6-12 months, measured against the first hold avoided.
• From 1-3 incidents a year with partial hold risk to zero.
• Discrepancies detected on the line, not at the packer client's plant. The pallet stays at the plant until the four records agree.
• And protection of the multi-year relationship with the packer client, which one poorly traced run can undo. Plus the BRCGS certification of both plants, which a single incident can put on the table.

Estimated payback of 6 to 12 months against the first hold avoided, protecting both the one-off incident and the multi-year relationship with the packer client. Estimate to validate with quality and general management.

And the fair question from the production manager

“Do we need all twelve cases before this works?” — no. The rings are built in layers and each case pays on its own; with the run startup and the curing oven you already have rings 1 and 2. Where the rings compare records, matching batch, curve and result is an anchored task where the best models drop below 1.5% error [1], and JIDOKA AI stops on a verifiable discrepancy, never on model doubt, leaving the call to a person. When the doubt is the model's, the case goes to a person; when the records disagree, the pallet stays.

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