Torta del Casar PDO — thistle rennet does not perform the same every day, and the creaminess tells you weeks too late.

The problem

Three variables that change every day, and a single cheesemaker holding them together.

A correct Torta del Casar PDO is not decided in the ripening room: it is decided in the vat. And vegetable thistle rennet introduces three variables that industrial animal rennet never had:

1. The day's thistle — which harvest, which drying, what the real cardosin concentration is. The technical sheets are indicative, but the truth of each infusion is only visible in the pH curve.
2. The day's milk — sheep on open pasture give milk that changes with the grass, with the rain and with the stage of lactation. The thistle and the milk have to get along for the coagulation to work.
3. The temperature of the dairy and of the curd table — a detail of a degree and a half in the vat moves the whole curve and the final creaminess.

The cheesemaker knows this. But adjusting three variables at the same time, batch after batch, through a long working day, is a thankless job. If the coagulation fails, the problem shows up weeks later in the ripening room — when nothing can be corrected. And nobody knows for sure whether it was the thistle, the milk, the temperature, or all three at once.

How it fits the IRIS system

iLEAN does not take judgment away from the cheesemaker — it gives them the data that today arrives weeks late.

The problem with thistle rennet is not ignorance: it is information living on islands (the pH probe on one side, the supplier's thistle delivery on another, the creaminess in the ripening room weeks later) that, by the time it comes together, is no longer any use for correcting anything. iLEAN acts as the putty that closes the loop between vat and ripening room, without asking you to change anything in the traditional process.

Edge captures the pH curve in the vat and the creaminess in the ripening room. Connect captures the thistle batch and the day's recipe. The agent cross-references everything and learns to anticipate. The cheesemaker decides and signs.

The iLEAN pieces applied to Cynara thistle vegetable rennet:

• iLEAN Edge in the vat — a local terminal that reads the pH curve (existing probe or new one), the temperature and the time to cutting. The curve is recorded batch by batch, without anyone having to write anything in a notebook. It works with no network: if the dairy loses WiFi, Edge keeps capturing.
• iLEAN Edge in the ripening room — a second terminal with AI vision that reads the final creaminess at the end of ripening: how the rind sinks when you touch the torta, the look of the cut, the estimated viscosity. It ties that result back to the vat batch.
• iLEAN Connect — captures the day's thistle batch (which harvest, which supplier, which drying date) whether it comes from the delivery note, from the supplier's WhatsApp or from the cheesemaker's sheet. And it captures the day's milk (which tank, which farm). It puts everything within the agent's reach at second zero.
• Agent — cross-references pH curve + thistle batch + the day's milk + final creaminess. It builds the dairy's historical pattern. When a new batch goes into the vat with a combination that has historically produced a firm torta, it warns the cheesemaker before cutting. The cheesemaker decides and signs; the system never touches the dose on its own.

See the full IRIS architecture →

Before and after

Thistle rennet from memory vs. thistle rennet with iLEAN Edge

Aspect	Classic control	With iLEAN Edge + Connect
Recording the pH curve	Handwritten note in a notebook, one-off reading	Continuous curve per batch, no notes needed
Traceability of the day's thistle	Delivery note filed away, never cross-referenced with the batch	Thistle batch tied to every vat
Link between vat and final creaminess	The cheesemaker's memory, weeks late	Pattern learned batch by batch
Anticipating a firm torta	Does not exist — you see it in the ripening room when it is already too late	Warning before cutting when the curve drifts
PDO file per batch	Rebuilt by hand	Automatic dossier with curve, thistle and creaminess
The cheesemaker's knowledge	Lives only inside them	Captured as a reusable pattern

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

• Artisan PDO dairy producing from a few dozen to a hundred tortas per working day, two working days a week.
• Edge pilot on the coagulation vat (probe + terminal + integration with the cheesemaker's notebook). First value expected within a few weeks: a pH curve available per batch and cross-referenced with ripening-room creaminess once the pilot batches come out.
• Expected reduction of firm tortas / out-of-spec PDO pieces of ≥ 30%, conservatively. Indicative payback between 5 and 9 months depending on volume and price per PDO unit.
• The hard lever is the batch held or adjusted in time: a firm torta cannot be sold as a Torta del Casar — and the prices are very different.

And the cheesemaker's reasonable doubt

“What if the AI gets the rennet dose wrong and ruins a batch for me?” — the AI does not touch the dose. It warns. Hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely learns the pattern between two time series (the pH curve and the final creaminess) and warns in advance, the best models brought error below 1.5% ^[1]. And even so: the agent proposes, the cheesemaker 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.