The job traveler, in the system at second zero

The problem

The only record of real progress is a sheet clamped to the part.

A large mold breaks down into hundreds of details. Each one passes through roughing, heat treatment, five-axis finishing, EDM, grinding and the bench, and each one carries its paper. The project lead needs to know every morning what share of the mold is genuinely done in order to defend — or correct — the delivery date, and what he actually has is a folder of sheets scattered across the machines. By the time the data reaches the system it is history: it supports billing, not decisions. And the margin notes that really matter ("roughing left more stock than planned", "drawing revision C is missing") never reach anyone. The result is a shop that works well and a schedule flying blind. The committed date gets defended with perceptions because the system has nothing better to offer.

• Every detail carries its traveler through roughing, heat treatment, five-axis finishing, EDM, grinding and the bench. The operator writes the actual hours and closes each operation by hand, and that sheet is the only place the truth lives.
• The project lead defends the tooling date every morning from a folder of sheets scattered across the machines. What the system holds was typed one to three days ago, and it feeds billing, not the schedule.
• The margin notes are the ones that matter — "roughing left more stock than planned", "drawing revision C missing" — and they die on paper because nobody types margins.
• The result is a shop that works well and a schedule flying blind. The committed date is argued with perceptions because nothing better exists.

How it fits the IRIS system

Connect in photo mode — the traveler stays, it just stops being the only copy.

Connect in photo mode, the cheapest capture mode to deploy because it changes nobody's habits.

It is the cheapest of the twelve to deploy because it changes nobody's habits: the machinist adds a photo to what he already does when he closes an operation. From that moment the detail exists in central memory with its hours, its revision and its handwritten issue, and the project lead sees the mold as the sum of its real details. The scheduling, the cost roll-up and the mold dossier all draw on the same capture, so one photo feeds three uses that today are fed by three separate transcriptions.

• The machinist photographs the job traveler with a phone or industrial tablet as each operation ends.
• A grounded LLM reads the document structure and extracts the fields: mold, detail, operation, drawing revision, hours and the handwritten issue.
• What was captured is summarized on screen and passes the human validation gate before entering central memory.
• Progress is tied to the project and available for scheduling, for cost roll-up and for the mold dossier.

The paper can stay if the shop wants it. It simply stops being the only carrier of the data.

See the full IRIS architecture →

Before and after

Today's traveler versus the captured traveler

Aspect	Today, on paper	With iLEAN Connect
Latency to the system	1-3 days, more over a weekend	Second zero
Mold progress	Eyeballed in the morning meeting	Sum of real details, per operation
Actual hours per detail	Typed later, rounded	Read from the sheet, tied to the detail
Handwritten issue	Dies on paper	Extracted and routed to whoever can act
Drawing revision on the sheet	Nobody cross-checks	Compared with the current revision
Machinist's habit	—	Unchanged: one photo

from 1-3 days of latency between "the operation is done" and "the system knows" → immediate. From project progress eyeballed in the morning meeting → real progress per detail. From handwritten issues dying on paper → signals that reach whoever can act.

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 4-9 months, depending on the number of live details in the shop.
• Transcription time recovered by the project lead and the office.
• From one to three days of latency down to second zero, detail by detail.
• Early detection of the details eating the project margin — the ones where roughing left stock or an operation ran twice the planned hours.

estimated payback 4-9 months depending on the number of live details, between recovered transcription time and early detection of details eating the project margin. *Estimate to validate*.

And the fair question from the production manager

"What if it misreads the hours a machinist wrote in a hurry?" — reading fields from a document whose structure is known is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: the summary reaches the bench tablet and a person confirms or corrects it before anything crosses into central memory, so a doubtful digit never becomes project data on its own.

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