A second objective reading

The problem

Grading is a judgment, and the judgment changes with the shift.

Two graders can score the same carcass differently; a mis-identified pack at shipping is a real risk with several product flows.

• Carcass grading is done by eye on a moving rail, under time pressure, and it directly sets what the carcass is worth.
• Two experienced graders can score the same carcass differently, and the same grader can drift across a long shift without anyone noticing.
• When a supplier or a toll client questions a grade, there is nothing to show beyond the grader's word, because no image of that carcass was kept.
• At the other end of the plant the same problem appears in another form: a pack loaded against the wrong delivery note, which with several product flows on one dock is a real risk rather than a theoretical one. Both cases share a root cause — a judgment made quickly, by eye, with nothing kept afterwards that would let anyone check it.

How it fits the IRIS system

Edge on the rail — the grader decides, the camera gives a second opinion.

Camera + CNN trained on the real grading code. A second camera at shipping cross-checks piece/pack against the delivery note.

It does not replace the grader and it is not designed to. A camera trained on your own grading code proposes a reading with its confidence on screen, and the grader confirms or overrides in the same movement they already make. The override is stored too, with its author, which is how the model learns where your plant and it disagree and where the disagreement is systematic rather than occasional.

See the full IRIS architecture →

Before and after

Grading today versus grading with a second reading

Aspect	Today	With iLEAN Edge
The grade	One subjective judgment	A judgment plus an objective reading
Drift across a long shift	Invisible	Visible as divergence
Between two graders	Two criteria	One shared reference
A grade questioned later	The grader's word	Image, reading and who confirmed
Pack against delivery note	Checked by eye at the dock	Cross-checked before loading
Who has the final word	The grader	The grader, unchanged

100% subjective grading → objective supporting reading.

Impact estimate

Impact estimate — to be validated with your numbers.

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.

• Estimated payback 5-12 months, depending on volume and on how much of your revenue moves with the grade.
• From fully subjective grading to a supporting reading that makes divergence measurable instead of anecdotal.
• A questioned grade gets answered with the carcass image and the reading taken at that moment, not with a recollection of a rail that moved on hours ago.
• And at dispatch, a pack that does not match its delivery note is caught before the truck leaves rather than at the customer's goods-in.

Estimated payback of 5-12 months. *Estimate to be validated*.

And the fair question from the production manager

“Is a camera going to grade better than a grader with twenty years on the rail?” — that is not the claim. The camera is consistent, the grader has judgment, and the two together are better than either alone. On a fixed rail with controlled lighting the reading is an anchored task where the best models drop below 1.5% error [1], and the system never posts a grade on its own: it proposes, the grader decides, and every override is kept.

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