The flagship system against size-color mix-ups

The problem

A mixed shipment is never a single mistake.

shipments with mixed-up size/color or wrong specification are this sub-sector's biggest reputational and contractual risk when serving large customer programs.

• A shipment with mixed-up size or color, or the wrong spec, is this sub-sector's biggest reputational and contractual risk when serving large customer programs.
• It rarely comes from one failure: a bundle guessed after its ticket was lost, a spec retyped at the coder, a leftover bundle after a changeover. Each one is small and each one is already covered by some control.
• Each link passes its own check, and together they put a size L in a carton labeled XL.
• On a corporate or government uniform program, one mislabeled shipment can cost the program's trust and the whole contract. And the penalty is usually written into the program agreement long before anyone reads it on the floor.

How it fits the IRIS system

Four rings that cross-check — the cases above, orchestrated.

Ring 1 (ERP) sets the spec; Ring 2 (coder) prints with zero typing via API; Ring 3 (Edge) cross-checks the print against the plan and JIDOKA AI blocks it if it doesn't match; Ring 4 (evidence pack) rides with every box. SMED AI keeps changeover speed up.

No single ring solves it; each exists today in some form and the mix-up still escapes. Adding more inspection at the end does not fix it either. The strength is the cross-check: when the ERP, the label and the camera stop agreeing, the carton stays, and SMED AI keeps changeovers fast while it does. Each ring is a case you can deploy and pay back on its own; the flagship is what they add up to.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes in workwear sewing

Ring	Which cases it draws on	What it closes
1 · ERP	Bundle ticket (1) + fabric receiving (7)	The spec and the dye lot each order is built from
2 · Coder	Label coder (8)	Size-color-lot labels printed with zero typing
3 · Edge	Sewing defects (9) + changeover (10)	The print checked against the plan; JIDOKA AI blocks on mismatch
4 · Evidence pack	Audit dossier (11)	Proof that rides with every box
Validation layer	Tablet validation (5)	A person behind every value in the ERP
Changeover speed in workwear sewing	SMED AI	Short runs without losing cadence

incidents/year with return or program penalty risk → zero. Discrepancies caught on the line, not at the customer.

Impact estimate

Impact estimate — to be validated with quality and plant management.

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, measured against the first incident avoided.
• Protection against contractual penalties and against losing an entire customer program. That is the risk that keeps a plant manager awake, not the cost of one return.
• Incidents per year with return or program-penalty risk: to zero. Each one avoided is measured against what the last one cost.
• Discrepancies caught on the line, not at the customer. And every carton leaves with the evidence of what was checked, ready if the customer ever asks.

protection against contractual penalties and losing an entire program, estimated 6-12 month payback against the first incident avoided. *Estimate to validate* with quality and plant management.

And the fair question from the production manager

“Do we have to deploy all twelve cases to get this?” — no, and planning it that way is the surest way never to start. The rings are built in layers and each case pays on its own. And JIDOKA AI stops on a verifiable discrepancy between sources — checking the printed label against the plan is an anchored comparison, where the best models drop below 1.5% error [1] — never on model uncertainty, which is escalated to a person.

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