4 coordinated rings against batch rejection

The problem

One rejected closure batch can stop a bottler's filling line.

A batch rejection by the bottling customer — leak, out-of-range torque, off-tolerance color — can stop their filling line and put a multi-year contract at risk. The active order defines mold, color, resin batch and destination customer.

• A batch rejection by the bottling customer — a leak, out-of-range removal torque, a color off tolerance — can stop their filling line and put a multi-year contract at risk.
• It rarely comes from one failure. It is a resin batch at the edge of spec, a purge cut short, a cavity running warm and a startup signed in a hurry, each acceptable on its own.
• The active order defines mold, color, resin batch and destination bottler. When nothing cross-checks the line against that order, the discrepancy is found at the bottler's incoming inspection.
• By then the batch is on their dock, the sorting and freight are yours, and the conversation is about the contract, not the closure. Even when the contract survives, the next renewal starts from a weaker position.

How it fits the IRIS system

Four coordinated rings — production order, vision, Edge and pallet dossier, orchestrated by Agents.

The piece-by-piece verdict from the existing vision system is automatically linked to the batch through the inspection-camera-to-ERP integration. Edge cameras validate mold and color before full-speed startup; if it doesn't match the order, JIDOKA AI blocks the startup. Every pallet ships with the dossier that cross-references the 3 previous rings and the quality sign-off. SMED AI speeds up changeovers so productivity isn't penalized.

No single ring solves the problem; each already exists in some form today and the bad batch still ships. The strength is in the cross-check: when the order, the vision verdicts, the startup images and the dossier stop agreeing, the pallet stays at the plant. It leaves when the discrepancy is explained and signed, not before, and the bottler never sees it. That is the whole promise: discrepancies stay a plant matter.

See the full IRIS architecture →

Before and after

The four rings against a closure batch rejection

Ring	Which cases it draws on	What it closes
1 · Production order	Mold startup photo (1) + resin receiving (7)	Which mold, color and resin batch make up the batch
2 · Vision	Inspection camera stitched to the ERP (8)	Piece-by-piece verdicts linked to the batch
3 · Edge at the press	Defective closure (9) + mold changeover (10)	Wrong mold or shade blocked before full speed
4 · Pallet dossier	Evidence pack (11) + quality sign-off	Every pallet ships with its cross-referenced file
Ring coordination	JIDOKA AI + SMED AI	Stops on mismatch without slowing changeovers
Human gate	Verification tablet (5)	No closure data reaches the ERP unchecked

Rejection risk caught by the customer → discrepancy caught on the line, before the pallet ships.

Impact estimate

Impact estimate — to be validated against your bottler contracts.

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 batch rejection avoided. One avoided rejection usually pays for several layers.
• Protection of the multi-year commercial relationship with the bottling customer, which is worth far more than the batch itself. Bottlers renew with the supplier who never stopped their line.
• The discrepancy is caught on the line, before the pallet ships, instead of at the bottler's incoming inspection. Sorting, freight and penalties simply do not happen.
• SMED AI keeps validation from penalizing changeovers, so the protection does not cost press productivity.

Payback 6-12 months · protects multi-year contracts. Protection against batch rejection and against the multi-year commercial relationship with the bottling customer. Estimated payback of 6 to 12 months against the first avoided rejection. Estimate to validate.

And the fair question from the production manager

“Do we have to deploy all twelve cases before this works?” — no, and framing it that way would be the way never to start. The rings are built in layers and each case pays on its own. Most plants start with one ring and add the next once the first has paid. Checking that order, verdicts and startup images agree is an anchored task, where the best models drop below 1.5% error [1]; with the mold startup photo and the vision stitching you already have rings 1 and 2. Edge and the dossier then close the loop.

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