JIDOKA AI: no startup until clean and validated

The problem

A recipe change is signed off from memory.

Switching from a standard bread flour to a specialty pastry flour requires cleaning and recalibration lasting 30 minutes to 2 hours, today signed off from memory by the mill supervisor.

• Switching the milling line from a standard flour to a high-strength specialty flour means cleaning and recalibrating: hoppers, roller mills, plansifters, spouts. Every mill has its own sequence, refined over years by the head miller.
• It takes 30 minutes to 2 hours, and today it ends with a sign-off by the mill supervisor based on memory, not on proof. Under time pressure, because the line is not producing while it is being cleaned.
• If residue of the previous flour stays in a sifter or a spout, the first tons of the new recipe carry it, and the W or the gluten of the batch drifts.
• When a customer complains about a specialty flour, the investigation finds a signature and no evidence of what the line looked like. The mill can neither prove the cleaning nor learn from what went wrong.

How it fits the IRIS system

Edge with JIDOKA AI — the line does not start without visual evidence and a signature.

Edge cameras at critical points (hopper, sifters, bagging spout) compare the current state against the "clean/ready" reference. iLEAN won't allow startup until all cameras give the OK and the quality manager signs off. JIDOKA AI.

The camera does not replace the quality manager's decision: it gives them evidence to decide on. What changes is not who signs, but what they sign. A signature backed by images of each critical point can be defended in front of a customer or an auditor.

See the full IRIS architecture →

Before and after

Recipe changeover today versus changeover validated by Edge

Aspect	Today	With iLEAN Edge
Sign-off after cleaning	From memory	Against reference images
Hopper, sifters, bagging spout	Checked if there is time	Each point verified by camera
Residue of the previous flour	Found in the first batch	Detected before startup
Startup permission	The supervisor's word	All cameras OK plus quality sign-off
Evidence for a complaint	A signature	Photos of every point, timestamped
Cross-contamination between recipes	Assumed low	Measured and documented

From a sign-off with no objective proof, to verified visual evidence at every point before startup.

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 6-11 months.
• Lower risk of cross-contamination between recipes, especially on high-value specialty flours. Those are the flours where a drift in W or gluten costs most.
• Fewer complaints because a batch did not match the customer's spec sheet after a changeover. The customer of a specialty flour notices before anybody else.
• And a startup backed by visual evidence at every critical point instead of a signature with nothing behind it.

Estimated payback of 6-11 months from lower cross-contamination risk between recipes and fewer spec-sheet complaints. Estimate to validate.

And the fair question from the production manager

"What if the camera blocks the restart for nothing and we lose an hour?" — that is the right objection, because the changeover is on the critical path. Comparing a point against its clean reference image is an anchored task, where the best models drop below 1.5% error [1]. And in case of doubt the system does not block silently: it shows the image to the quality manager, who decides on the evidence. The line stops on a verified mismatch, not on the model's uncertainty.

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