Seamless traceability, from sack to pack

The problem

A recall is almost never caused by a single failure.

any gap — moisture, allergen, untraced cleaning — can escalate into a product recall or an audit non-conformity, with a team too small to watch every point by hand.

• It is the concatenation: an origin lot typed in a hurry at the dock, a drying run whose moisture nobody logged, a changeover signed without evidence and a formula version that was never recorded.
• Each link on its own looks acceptable, and each one passes its own control. Together they produce the batch you cannot defend when the question comes.
• That is why isolated controls are not enough: they have to cross-check each other, because what gives a gap away is two independent sources ceasing to agree about the same batch.
• And the team that would have to hold every point by hand is the same lean quality team that also runs the audits, the specification sheets, the customer complaints and the supplier approvals.

How it fits the IRIS system

Four rings that cross-check — it is all of the above, orchestrated.

Connect captures every signal; Edge visually monitors critical points; Agents coordinates the auditable dossier; JIDOKA AI blocks startup on anomalies; SMED AI speeds up formula changeovers between batches.

No single ring solves the problem: each one already exists today in some form and the gap still appears. The strength is in the cross-check, because a traceability chain is only as good as its weakest link, and this is what makes the weak link announce itself instead of waiting to be discovered by a customer.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · Origin	Intake (7) + supplier alert (3)	Which grower, harvest and lot the batch is made of
2 · Process	Batch startup (1) + dryer panel (2) + lab (6)	Formula, moisture, color and heat, batch by batch
3 · Allergen	Blend validation (5) + allergen cleaning (10)	What the batch declares and what the equipment held before
4 · Evidence	Encoder (8) + packaging (9) + evidence pack (11)	What was printed, what shipped and the dossier behind it
Cross-check	JIDOKA AI over all four	Startup blocked when two sources disagree
Changeovers	SMED AI between batches	Formula changes without the evidence gap

traceability rebuilt by hand under pressure → live traceability, queryable by batch at any moment.

Impact estimate

Estimated impact — to validate with your own 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.

• No standalone payback figure is claimed for the whole system: each layer pays on its own, and the combined number depends on which order you deploy them in.
• Non-conformity and recall risk reduced by an order of magnitude, which is the pain that defines a plant of this size.
• Traceability queryable by batch at any moment, instead of a reconstruction exercise under pressure.
• And the quality team freed from transcription work, which is the capacity you actually needed to certify the next standard.

non-conformity/recall risk reduced by an order of magnitude, quality team freed from transcription work. *Estimate to validate*.

And the fair question from the production manager

"Do we have to deploy all twelve cases to get this?" — no, and framing it that way is the way never to start. The rings are built in layers and each case pays on its own from the first month. With intake and batch startup you already have ring 1 and half of ring 2, which is where most of the missing traceability sits. Within each ring the underlying tasks stay anchored, where the best models drop below 1.5% error [1], with a person on the gate.

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