A PDO wheel is not one link in a batch — it is a unique asset with its own history.

The problem

The batch does not tell what each wheel lived through in the cave.

A PDO cheese is made in hours but earned over months. Two wheels born from the same vat of curd end up on different shelves, with different humidity, with different cheesemakers, with brushings out of sync. When one wheel causes a problem in the market — an out-of-range analysis at a demanding retailer, a complaint from a gourmet customer, an internal alert — the classic system offers two bad options:

1. Put the entire pasteurizer batch on hold as a precaution. For a vat of 200 wheels, that means freezing significant market value while the counter-analyses are run. If everything comes back clean, you have absorbed the financial cost and blocked a sales channel for nothing.
2. Work out by hand which wheel was which. The vat of origin is in the ERP, yes, but the position in the cave, the turnings and the individual brushing live in spreadsheets, in a notebook and in the cheesemaker's head. Reconstructing the itinerary takes days — the supermarket wanted an answer by tomorrow.

And the problem is not only reactive. It is also proactive: the premium B2B customer pays for the exact provenance of their wheel — which herd, which vat, which cave, which week — not for a generic PDO label. Without wheel-by-wheel traceability, the brand premium evaporates.

How it fits the IRIS system

iLEAN Tracer is the filler that binds the vat to the cave to the rind.

The problem is not a lack of information: it is information split into islands — ERP with the vat, spreadsheet with the position in the cave, the cheesemaker's notebook with the turnings, a humidity sensor with no context. iLEAN Tracer does not ask you to change your ERP, your MES, or the way you make cheese. It sits on top and seals the cracks.

Edge marks the wheel at moulding. Connect captures every cave operation wherever the data lives (spreadsheet, notebook, IoT). The agent cross-references recipe + cave + sanitary results and assembles the file. The person signs off — never the other way round.

The iLEAN pieces applied to wheel-by-wheel traceability:

• Edge in the moulding room — a vision-equipped terminal that sees each freshly moulded wheel and triggers the marking on the rind (casein seal, laser accepted by the Regulatory Council, or food-grade printing) synchronised with the assigned code. No network available is not a problem: the Edge keeps reading and marking with its own light.
• Connect in the ripening cave — the moulding recipe, the humidity/temperature probe, the cheesemaker's turning notebook and the result of every intermediate analysis enter the system through whichever channel you already use (a photo of the notebook, an IoT reading, voice dictation, a spreadsheet forgotten in a shared folder). At second zero, nothing is lost.
• File agent — cross-references the vat of origin, the cave itinerary, the cheesemaker's interventions and the sanitary analyses against the wheel's code. When the question arrives — from the B2B customer, from the Regulatory Council, from the retailer — the per-wheel dossier is already built. If an incident appears, the agent proposes the exact list of wheels with the same history; the person validates and signs off.

See the full IRIS architecture →

Before and after

Pasteurizer-batch traceability vs. wheel-by-wheel traceability

Aspect	Pasteurizer-batch traceability	With iLEAN Tracer, wheel by wheel
Traceable unit	Vat of curd (dozens or hundreds of wheels)	Each PDO wheel, code on the rind itself
Ripening history	Cave average, not individual	Position, turnings, brushing and humidity per wheel
Out-of-range analysis on one wheel	Put the whole batch on hold as a precaution	Separate the wheel with that exact history
Dossier for the Regulatory Council	Rebuilt by hand from several sources	Automatic per-wheel file, ready to go
Premium B2B margin	Hard to justify — generic label	Exact provenance — vat, cave, moulding date
Operation without network in the cave	n/a	Edge keeps reading and recording locally

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized PDO dairy, 2-3 ripening caves, marking in the moulding room + Edge reader in one cave zone as a pilot.
• First value expected within a few weeks: a per-wheel dossier working for the marked cohort, demonstrable to the Regulatory Council and to the premium retailer.
• Indicative payback between 4 and 9 months, depending on the average historical cost of holds triggered by lab analyses and on the weight of the premium B2B channel in your mix.
• Hard levers: wheels not put on hold unnecessarily ≥ 30% of past cases; premium B2B margin ≥ 30% over channels that today buy as generic; hours saved rebuilding the audit file by hand.

And the quality manager's reasonable doubt

«What if the AI gets it wrong and puts the wrong wheel in the file?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualises a piece of data (reading the code on the rind and cross-referencing it with the cave itinerary), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: the agent proposes, the person signs off. The three safety rings exist precisely for this.

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