From raw material to pallet, four rings that verify each other

The problem

The pain is coordinating short-lived materials without losing a batch or the certificate.

Coordinating raw materials with short shelf life and specific storage needs against the production rhythm, without it costing a nonconforming batch or the organic certification that underpins exports: that's this plant's capital pain. Raw-material receiving gets validated by lot and expiry from minute one, via Connect. No expired or poorly traced ingredient enters a reactor unnoticed.

• Raw materials with short shelf life and specific storage needs — live strains, plant extracts, cold-chain inputs — have to meet a production rhythm that does not wait for a late truck or a slow lab result.
• Each control exists in some form: receiving, fermentation, cleaning, coding, release. Each one on its own looks acceptable.
• But a strain near expiry, a fermentation curve slightly off, a cleaning signed in a hurry and a hand-typed expiry add up to a nonconforming batch that passes every individual check and only fails in the grower's field, weeks later.
• And on a certified line, that batch does not only cost itself: it can cost the organic certification that underpins exports.

How it fits the IRIS system

Four rings that verify each other — the previous eleven cases, orchestrated with JIDOKA AI and SMED AI.

The reactor only starts each batch after visual cleaning confirmation via Edge, and the coder prints batch and expiry straight from the active work order, no typing.

No ring is enough alone, and each already exists in part today. The strength is in the cross-check: when receiving, fermentation, cleaning and coding stop agreeing about a batch, JIDOKA AI stops it before the pallet, and SMED AI keeps the changeovers from eating the productivity that safety would otherwise cost.

See the full IRIS architecture →

Before and after

Ring by ring, from strain receiving to released pallet

Ring	Cases it draws on	What it guarantees
1 · Receiving	Delivery note (7) + supplier alert (3)	No expired or poorly traced strain enters a fermenter
2 · Process	Work order (1) + fermenter panel (2) + lab sheet (6)	Formula, curve and bioassay tied to each batch
3 · Line	Organic cleaning (10) + container vision (9)	Clean vessel before organic batches, sound containers
4 · Release	Coder (8) + tablet gate (5) + evidence pack (11)	Correct batch and expiry, signed, with its dossier
Cross-check in biosolutions	JIDOKA AI	The batch stops before the pallet if rings disagree
Changeovers in biosolutions	SMED AI	Productivity kept while every check runs

Every batch ships with its auditable dossier cross-checking the three previous rings. If something doesn't match, JIDOKA AI stops the batch before the pallet, and SMED AI keeps productivity up during changeovers. Estimated payback 6 to 12 months against the first incident avoided.

Impact estimate

Impact estimate — set against one serious certification incident.

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 6-12 months against the first incident avoided.
• Protection against loss of organic certification and against nonconforming batches reaching growers and distributors, where the cost is counted in reputation as well as money.
• Every batch ships with an auditable dossier that cross-checks the three previous rings, so a complaint is answered with evidence, not reconstruction, and the containment covers one batch instead of a season.
• And changeovers between organic and conventional batches stay productive, because SMED AI organizes them while the checks run, instead of the checks being skipped to save time.

Protects against loss of organic certification and nonconforming batches · estimated payback 6-12 months against the first incident avoided.

And the fair question from the production manager

“Do we have to deploy all twelve to get here?” — no, and planning it that way would be the surest way never to start. The rings are built in layers and each case pays on its own. When the rings cross-check, comparing a batch's data across sources is an anchored task where the best models drop below 1.5% error [1]; JIDOKA AI stops only on a verifiable discrepancy, and a person decides. Start with receiving and the work order, and the first ring is working within weeks.

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