Continuous audit of the Comté AOP cahier des charges — the inspection does not catch you out, you are already inside.

The problem

Meeting the cahier des charges is not an exam, it is a permanent state.

Comté AOP is a cheese built in three places, on three different clocks: the producer's farm, the fruitière, and the affineur's cave. The cahier des charges ties all three to a single promise of origin and quality. And each of them works with its own paperwork:

1. The farm — feeding sheet, breed declaration, milk batch. Often on paper, or in the producer's own ERP that does not talk to the fruitière.
2. The fruitière — vat record, pH, rennet, temperature; it may come from the MES or from a notebook — and the copper vat, almost always from the notebook.
3. The cave d'affinage — temperature, humidity, turning and morge-brushing frequency, recorded room by room. A notebook or an ancient BMS.

The quality manager knows all this, but cannot be in three places at once. When the CIGC inspection arrives, what gets measured is not the promise, it is the evidence. And rebuilding a dossier per meule after the fact is what costs three shifts of hunting for paperwork. An audit finding is not a manufacturing failure: it is an evidence failure.

How it fits the IRIS system

iLEAN Brain does not add a fourth system — it seals the cracks between the three you already have.

The problem with meeting the cahier des charges is not a lack of information: it is information that lives in islands and only gets cross-referenced when the inspection comes. iLEAN acts as the putty that fills those gaps, without asking you to change your fruitière system, the producer's ERP, or the affineur's traceability.

Connect captures the milk, the vat and the cave. Brain cross-references them against the cahier des charges continuously. If something drifts, it flags it to the quality manager — not at 10 p.m., but at the moment. The person signs — never the other way round.

The iLEAN pieces applied to continuous audit of Comté AOP:

• Connect — captures, through whatever channel, the milk driver's collection sheets, the farms' feeding declarations, the vat parameters (integration or a photo of the panel), and the cave records. It also captures what arrives from outside (a rule change from the CIGC, a notice from the certification body) at second zero, without anyone forwarding anything.
• Brain (Agents + Central) — the brain that cross-references the three sources against the cahier des charges in real time. If a feeding sheet mentions something the cahier does not allow, if the vat pH has drifted out of range, if the cave has sat outside the humidity band for several hours, the agent flags it and proposes an action. It does not send an email at 10 p.m.: it alerts the quality manager through whatever channel they use, with the context attached.
• Dossier per meule — the natural by-product of continuous audit: when the inspector arrives, every wheel has its complete dossier, with its three sources cross-referenced and its deviations already handled. Zero manual reconstruction.

See the full IRIS architecture →

Before and after

Reactive audit vs. continuous audit with iLEAN Brain

Aspect	Notebook + ERP + cave notebook	With iLEAN Brain (Connect + Agents)
Milk-vat-cave cross-reference	Manual, whenever the inspector asks	Continuous, at second zero
Detecting a deviation from the cahier	At the inspection, weeks later	At the moment, straight to the quality manager
Producer feeding sheets	On paper, at risk of being lost	Captured by Connect — photo, voice or integration
A CIGC rule change	Arrives by email, gets printed, gets forgotten	Captured at second zero, replicated into the agent's rules
Dossier per meule for the inspection	Rebuilt by hand, three shifts of work	Ready, with its evidence, in one click
A claim from the affineur	Go find the cave notebook	An answer cross-referencing all three sources, within hours

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 fruitière. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Mid-sized fruitière, 15-30 milk producers, 1-2 vats a day, 1-2 active caves d'affinage.
• Brain pilot with integration of the vat MES, Connect capture of producer sheets and cave records, and mapping of the cahier des charges into agent rules. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency and average cost of the audit findings documented over recent years.
• Expected reduction in hours spent rebuilding dossiers by hand ≥ 30%, with a dossier per meule ready for the inspection.
• The hard lever is a single cave lot not downgraded: given what a long-ripened Comté cave is worth, one is enough to pay for the pilot.

And the quality manager's reasonable doubt

“What if the AI misreads the cahier des charges and lets a deviation through?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system to another (reading a sheet, comparing a range against the cahier), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN flags the deviation and the person signs. The three safety rings exist precisely for this.

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