No batch is released without a human having seen and signed it

The problem

The data is typed by whoever did not look at the part.

Today batch release is signed on paper and typed later, or typed straight into a system whose screen is slow and which the operator cannot reach from their workstation. Both options produce the same result: data arriving late and, frequently, entered by someone who never looked at the part. A batch released with the leak test signed by nobody in particular is the hole a containment campaign escapes through. When the customer asks who decided that part could ship, the answer has to exist and it has to have a name. The opposite mistake is just as expensive: ask the operator to fill in one more form and the form gets filled in badly or not at all. Any validation gate that takes time away from the line eventually gets bypassed.

• Today release is signed on paper and typed later, or typed into a system whose screen is slow and which the operator cannot reach from their station.
• Both options produce the same result: data arriving late and, often, entered by somebody who was not the person who looked at the part.
• A batch released with the leak test not really signed by anybody is the hole a containment campaign escapes through.
• When the customer asks who released it and on what, the answer has to be a name and evidence, not a checkbox.

How it fits the IRIS system

Connect with early human verification — two taps at the cell.

Connect in early-human-verification mode. This is the IRIS component that guarantees the master does not get contaminated and that every data point has a named owner. Step 1 — A summary, not a form. The industrial tablet sits at the cell on an articulated arm and asks you to fill in nothing: it shows a visual summary of what iLEAN already captured by any route — camera, panel, voice, parser — with the leak curve drawn and any dimension near its limit highlighted. Step 2 — Two taps. Confirm or correct. Nothing more. The cost to the line is seconds, not minutes. Step 3 — Only then does it cross. The data enters the MES with the signature of whoever validated it and a timestamp. Step 4 — The correction is learned too. When the operator corrects something, what changed and who changed it are stored: that leaves the trace of the decision and sharpens the system's judgement for the next part.

It is the piece that guarantees the master is not contaminated and that every datum has a responsible person by name. Without it, the other eleven do not get board approval: nobody signs off on AI writing into the system a release depends on.

See the full IRIS architecture →

Before and after

Today's release versus the signed release

Aspect	Today	With iLEAN
When the data enters	Hours later	In seconds, at the station
Who appears as the author	Whoever types it	Whoever looked at the part
What the signature proves	That somebody signed	Which result and which parameters they saw
Operator's time	Walk to a terminal	Two taps on the tablet
When the customer asks	A checkbox	A name and its evidence
Who writes to the master	A person, typing	A person, confirming

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.

• Estimated payback 4-9 months.
• The dominant value here is not productivity, it is risk: every release is signed with its evidence attached.
• Which is exactly what gets asked for when a field complaint lands and you have to show who released and on what.
• And authorship moves from whoever types the data to whoever looked at the part.

Estimated payback of 4 to 9 months, *estimate to validate*. The dominant value here is risk rather than productivity: every release ends up signed with its evidence attached, which is exactly what has to be shown when somebody asks why a specific part left the plant.

And the fair question from the production manager

«If a person signs in the end, what is the AI saving?» — the work is not in deciding, it is in assembling. Today somebody would have to go to the terminal, find the batch and type the result. With the verification gate it all arrives assembled in a summary and they only confirm or correct. And every correction is stored: the system learns where its extraction fails.

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