First article validated

The problem

Two pressures meet at the first article, and rigor is the one that loses.

The pressure to shorten changeover time competes directly with validation rigor, and time pressure wins more often than anyone admits out loud. A badly validated first article can drag an entire lot out of spec.

• Every mold change and every color change ends the same way: somebody looks at the first part and decides whether the run can go.
• That decision happens while the press is stopped, with the changeover sitting on the shift's critical path and everybody waiting.
• Time pressure wins more often than anyone says out loud, and the record of the decision is a signature, not a measurement.
• A badly validated first article does not stay a single bad part: it drags the whole lot out of spec, and the lot is already boxed by the time anyone finds out.

How it fits the IRIS system

Edge plus JIDOKA AI — no full run until there is evidence and a sign-off.

Edge + JIDOKA AI. Industrial cameras with CNNs trained to compare the first article against the approved master sample. Full-run start is blocked until OK + quality lead sign-off.

The camera does not decide alone. It compares the first article against the approved reference state point by point and hands the quality lead something to sign on, and the full run stays blocked until both the OK and the signature exist. What changes is not who decides, but what they decide on.

See the full IRIS architecture →

Before and after

Today's first article versus the validated first article

Aspect	Today	With iLEAN Edge
The first-article decision	An eyeball under time pressure	Compared against the approved master sample
Evidence behind the decision	A signature on a form	Images with every point checked
Start of the full run	Whenever the setter says go	Blocked until OK plus sign-off
A run started on a bad first article	The whole lot at risk	Cannot start
The changeover clock	Rigor traded away for minutes	Seconds added, rigor kept
Change on a shift with no quality technician	Signed anyway	Same criterion, remote sign-off

eyeballed validation under time pressure → objective visual-evidence validation, with no changeover speed given up.

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.

• Payback tied to how often you change mold or color — the brief deliberately gives no month range here, because the return is counted per changeover and not per calendar.
• What it protects is the whole lot after every change, which is the unit of loss in a molding shop.
• Changeover rigor stops competing with changeover speed, because the check costs seconds and the argument costs minutes.
• Bring your changes per week and the average lot size and the range comes out of your own numbers, not ours.

protects the entire lot after every changeover, payback depending on mold/color change frequency. Estimate to validate.

And the fair question from the production manager

"What if the camera blocks a run for no reason?" — the objection is the right one, because the changeover is on the critical path. The comparison is made against the reference state of that exact part number, and a doubt does not become a silent block: it escalates to the quality lead, who signs on the evidence. It stops on a verifiable discrepancy, not on model uncertainty — which is the line this matrix draws everywhere, including the anchored tasks where the best models drop below 1.5% error [1].

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