Iberian breed purity verification 100/75/50 — the vet's eye, extended to every animal.

The problem

The inspecting vet gets a proper look at one animal in ten. Vision looks at all ten.

Breed purity verification at the slaughterhouse is the last chance to reconcile what the herdbook says with what the animal shows, before the designation mark goes onto the carcass. The inspecting vet has the judgment and the experience, but not the time:

1. The line rate — the slaughterhouse cadence does not allow a thorough look at the coat, hoof and profile of every animal. Inspection is done on a sample basis.
2. The pedigree lives elsewhere — on the farm, on paper, in another system. The inspecting vet does not have it on screen as the carcass goes by.
3. The knowledge takes a career to build — a veteran vet sees what a junior does not. The day they retire, it goes with them.

When the designation leaving the slaughterhouse does not match the animal's real purity, the error survives all the way to the seal on the cured ham two years later — and by then it is a nonconformity from the certification body or, worse, a recall. 99% of the time the vet gets it right. It is that 1% that weighs.

How it fits the IRIS system

iLEAN does not replace the vet — it gives them a second look at every animal.

The problem with purity verification is not a lack of information: it is that the information (the animal's morphological traits, the herdbook, the ear tag, the farmer's logbook) lives on islands, and the line rate does not allow cross-referencing it on every animal. iLEAN acts as the putty that stitches that information together on the line, without asking you to change the slaughterhouse MES or the herdbook.

Vision reads the carcass on the line. Connect cross-references the pedigree and the ear tag. The agent proposes a purity and the reasons for it. The vet signs — never the other way round.

The iLEAN pieces applied to breed purity verification 100/75/50:

• iLEAN Vision (Edge + CNN) — an AI vision terminal on the carcass line. It reads coat, hoof, head profile and ham conformation on every animal, not on a sample. It runs locally: it does not depend on the network to record and propose.
• Connect — cross-references the visual data with the Iberian breed herdbook, with RIBER, with the animal's ear tag and with the farm record. It also captures the veteran vet's judgment when they dictate on a borderline case, and leaves it as in-house data for the ones that follow.
• Agent — proposes a purity (100/75/50) with a confidence level and the justifying traits. If there is a discrepancy between the pedigree and the morphology, it flags it. The carcass is only marked once the inspecting vet signs.

See the full IRIS architecture →

Before and after

Sample-based verification vs. verification assisted with iLEAN Vision

Aspect	Traditional verification	With iLEAN Vision (Edge + CNN + agent)
Coverage per carcass	Sampling (human capacity)	100% of the animals
Cross-check with the pedigree	Separate, in another system	In the proposal, on the inspector's screen
Justifying traits	Mental judgment, no trace	Photo and extracted traits, anchored to the ear tag
The veteran's knowledge	In their head	Captured as model criteria
Discrepancy detection	A later audit or a complaint	On the line, before the carcass is marked
File for the certifier (ENAC)	Manual rebuild	Dossier per carcass with image and signature

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

• Iberian slaughterhouse with two carcass lines, multiple designations (100/75/50), purity verification done today on a sample basis by the inspecting vet.
• Vision pilot on one line (terminal + camera over the carcass + integration with the MES and the herdbook). First value expected within a few weeks: 100% coverage with a reasoned proposal per carcass.
• Indicative payback between 4 and 9 months, depending on how often designation nonconformities come in from the certification body and on the cost of the inspecting-vet hours spent today rebuilding the file.
• Expected reduction in discrepancies between designation and morphology ≥ 30%; the rest translates into fewer recalls downstream and a better position with the Ministry of Agriculture.

And the inspecting vet's reasonable doubt

“What if the AI proposes the wrong purity and a junior signs it without thinking?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely reads traits and checks them against the pedigree, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: the agent proposes with visible reasons and the vet signs. The system requires the inspector to see the coat, the hoof and the conformation behind the proposal — not to accept it blindly. The three safety rings exist precisely for this.

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