The most dangerous moment of the day closes today with a signature

The problem

The most dangerous moment of the day closes with a signature and no evidence.

With dozens of references sharing the same kettles, forming machines and breading lines, and a matrix including milk, gluten, egg, fish, crustaceans and molluscs, critical changeovers happen several times a day, every day. The signature that closes them is honest and expert. The problem is not the signature: it is that it leaves no objective evidence behind. If there is a complaint tomorrow, what exists is a signed box. And on a clean label product, without preservatives, there is no chemical safety net covering anything. Process hygiene and allergen control carry the whole weight, which makes the tolerance margin tighter than a conventional competitor's.

• A run of prawn croquettes — crustacean — finishes and a spinach and cheese run comes in. The cleaning between the two is the only thing separating a correct product from an allergen incident.
• With dozens of references sharing the same kettles, forming machines and breading lines, and a matrix that includes milk, gluten, egg, fish, crustaceans and molluscs, that transition happens several times a day, every day.
• The signature that closes it is honest and expert. The problem is not the signature: it is that it leaves no objective evidence behind. If there is a complaint tomorrow, what exists is a signed box.
• And on a clean label product, without preservatives, there is no chemical safety net: process hygiene and allergen control carry the whole weight, which makes the margin tighter than a conventional competitor's.

How it fits the IRIS system

Edge with JIDOKA AI — the line does not start without conformity at every point and a signature.

Edge + JIDOKA AI.

The cameras do not decide alone: each one compares its critical point against the clean reference state for the recipe about to run, and it is the quality manager who signs on those images. What changes is not who decides, but what they decide on — evidence instead of memory.

• Fixed cameras on the critical points the plant defines: breadcrumb hopper, forming machine, belt, pre-fryer.
• Each camera compares the current state against the "clean and ready to run this recipe" reference state, with its own model and criterion: breadcrumb residue, dough stuck on the former, carry-over in the oil.
• JIDOKA AI prevents start-up until every point clears.
• The quality manager sees the images with their verdict and signs on evidence rather than on memory: the system does not replace the signature, it supports it.
• All images are archived and feed the evidence pack of case 11.

See the full IRIS architecture →

Before and after

Today's changeover versus the verified changeover

Aspect	Today	With iLEAN Edge
What closes the changeover	A signature on a form	Four critical points checked against reference, then a signature
Breadcrumb residue in the hopper	Depends on who looked	Compared against the clean state
Dough stuck on the former	Found on the next run, or not	Blocks start-up
Carry-over in the pre-fryer oil	Invisible	Its own model and criterion
Evidence after a complaint	A signed box	Images per point, archived automatically
Start-up under shift pressure	Depends on the workload	Blocked until every point clears

Impact estimate

Impact estimate — to be validated against your non-conformity history.

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.

• Payback measured against a single allergen incident: recall, distributor penalty and reputational damage to a premium product. We give no month range here on purpose; the number depends on how many critical changeovers you run per week.
• Cross-contamination risk stops depending on the shift's workload: the start-up block is automatic, not a matter of who was in a hurry.
• Every changeover leaves images per critical point, archived and tied to the batch, which is the evidence the audit case then organizes.
• And SMED AI, in the flagship, makes sure that rigor does not penalize productivity: the sequence gets faster while the verification stays.

the return is measured against the cost of a single allergen incident — recall, distributor penalty and reputational damage to a premium product. The payback depends on the number of critical changeovers per week. *Estimate to be validated against the non-conformity history*.

And the fair question from the production manager

“What if the cameras block the line for no reason?” — that is the right objection, because the changeover sits on the shift's critical path. Each point is compared against the reference state for that specific recipe, and comparing a fixed scene against a trained reference is an anchored task, where the best models drop below 1.5% error [1]. In case of doubt the system does not block silently: it escalates the image to the quality manager, who signs or sends the point back for cleaning. It stops on a visible discrepancy, not on model uncertainty.

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