The code that identifies the unit cannot depend on a keystroke

The problem

One wrong keystroke and a whole run ships as unidentifiable units.

At every model changeover somebody keys the variant, the engineering revision and the marking format into the marker's panel. One wrong keystroke and a whole run ships marked with the previous revision — which is, technically, an unidentifiable unit. The same happens at the other end: the end-of-line test bench keeps its results on its own disk, and somebody exports them by hand when asked. The unit's birth history stays incomplete until that manual work is done. Proper integration of these machines has been on engineering's wish list for years and never quite reaches the top, because each one speaks a different protocol and the project competes with product launches. Meanwhile manual keying stays the standard method at the single most critical point of the whole traceability chain.

• The laser marker is the last piece of equipment to touch the product: it engraves the data matrix code that identifies that airbag module or pretensioner for the rest of its life. It is a modern machine with an API — and at every model changeover somebody keys the variant, the engineering revision and the marking format into its panel.
• One wrong keystroke and a whole run ships marked with the previous revision, which is, technically, a set of unidentifiable units. That is the scenario that turns a suspect lot into a broad recall campaign.
• At the other end, the end-of-line test bench keeps its verdicts and measurements on its own disk, and somebody exports them by hand when asked. The unit's birth history stays incomplete until that manual work is done.
• Proper integration has been on engineering's wish list for years and never reaches the top: each machine speaks a different protocol and the project competes with product launches. So manual keying stays the standard method at the single most critical point of the traceability chain.

How it fits the IRIS system

Connect stitching two modern systems through APIs — neither machine is replaced or touched inside.

Connect stitching two modern systems through APIs.

When the MES marks the start of an order, it publishes variant, current revision and marking format to the marker; in the other direction, the test bench publishes its verdict and its measurements. Everything is anchored to the serial number in the traceability master, so the unit's birth history composes itself in real time and a model changeover becomes a tap in the system instead of a typing job on a machine panel.

• Replacing neither machine and touching neither one's internal logic.
• When the MES marks the start of an order, it automatically publishes variant, current revision and marking format to the marker.
• In the other direction, the test bench publishes its verdict and its measurements.
• Everything is anchored to the serial number in the traceability master, composing the unit's birth history in real time.
• A model changeover becomes a tap in the system, not a typing job on the machine's panel.

See the full IRIS architecture →

Before and after

The isolated marker versus the stitched marker

Aspect	Today	With iLEAN Connect
Variant, revision and format at changeover	Keyed by hand on the panel	Published by the live order
Run marked with the previous revision	A permanent risk	Eliminated by construction
Test bench verdict and measurements	On the bench's disk, exported on request	In the unit's history in real time
Unit's birth history	Complete after manual export	Complete as the unit leaves the line
Model changeover on the marker	A typing job	A tap in the system
Either machine	—	Neither replaced, neither touched inside

from keying errors in marking (with whole runs to re-identify) to zero human-origin error · from test results exported by hand to a complete birth history in real time · from an out-of-date revision on the panel to always the current one.

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 3-6 months, the fastest of the twelve — starting by confirming the current marker's API is available.
• A drastic reduction in the risk of a misidentified unit, which is the scenario that turns a suspect lot into a broad recall campaign.
• Test results in the unit's history in real time instead of exported by hand when somebody asks.
• The revision on the marker is always the current one, on every line and at every changeover.

fast return, 3-6 months; a drastic reduction in the risk of a misidentified unit — the scenario that turns a suspect lot into a broad recall campaign. *Estimate to validate*, starting by confirming the current marker's API is available.

And the fair question from the production manager

"What if the marker engraves something other than what the order published?" — the publication itself is deterministic, API to API, with no model in between. Where a model does intervene — reading the bench's verdict or confirming what was engraved from its return message — the task is anchored against a known serial number, where the best models drop below 1.5% error [1]. And the bench cross-checks the DMC it reads against the one the order expected, unit by unit.

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