The heat sheet, digitized at second zero

The problem

The most important document in the plant is still a sheet of paper.

Traceability for a brake disc or a caliper is anchored to the heat number. While the sheet is paper, that number only exists on the shop floor. The planner does not know which heats are running clean, quality cannot cross chemistry with defects, and logistics cannot bound a batch if a question appears tomorrow. When it finally gets transcribed, it gets transcribed late and sometimes wrong. A mistyped figure on a heat sheet is a quality investigation that starts from false data, and on safety-critical parts that multiplies the scope of any later containment. On top of that, the richest information on the sheet is the operator's handwritten notes — what they saw, what they adjusted, what looked odd — and those notes never reach any system at all.

• The heat sheet carries the charge makeup, the tapping and pouring temperature, the magnesium treatment time, the inoculant and the heat number that will follow every brake disc and caliper for its whole life.
• It is closed by hand at the furnace and typed at the end of the shift, or the next morning. Four to eight hours in which the heat exists physically but in no system. Over a weekend, two days.
• A mistyped magnesium percentage or a heat number off by one digit starts a quality investigation from false data, and on safety parts that widens any later containment to heats that were never suspect.
• The melter's margin notes — a slow tap, a cold ladle, an extra addition — are the richest information on the sheet, and they reach nobody. When the defect shows up two weeks later, the note that explained it is in a folder.

How it fits the IRIS system

Connect in photo mode — the melter keeps closing the sheet by hand.

IRIS handles this with Connect in "photo of the analog" mode, built for whatever is still paper on the floor:

Nothing changes at the furnace: the sheet is still the plant's own form, filled in the way it always was. What changes is that one photo when it is closed makes the heat exist in the central system at second zero, with the melting supervisor confirming what was read before it crosses into the ERP. It is the simplest of the twelve to deploy and the one the melters accept first, because it asks nothing of them.

• The operator photographs the sheet with a phone or industrial tablet when closing it. No habit change.
• A grounded model recognizes the plant's own form layout and extracts the fields: heat number, alloy, target chemistry, temperature, treatment time, furnace, shift and the handwritten notes.
• What is extracted appears on the line tablet next to the original photo for early human verification: confirmed or corrected in two taps.
• Only then does the data enter central memory and become available to the rest of the system: the furnace curve, the lab result, the defects seen in line.

Humans in command: iLEAN structures and proposes, the person signs.

See the full IRIS architecture →

Before and after

Today's heat sheet versus the captured heat sheet

Aspect	Today, on paper	With iLEAN Connect
Metal-to-data latency	4-8 h, longer over a weekend	Second zero
Heat number in the system	Typed at end of shift	Extracted and confirmed at the furnace
Magnesium treatment time	A figure in a folder	Queryable by heat and by furnace
Melter's handwritten notes	Never leave the sheet	Searchable text tied to the heat
Typing errors	Silent until the investigation	Verified against the source photo
Melter's habit	—	Unchanged: one photo

from 4-8 h of latency between metal and data to second zero · from handwritten notes nobody reads to text searchable by heat · from manual end-of-shift transcription to none · from silent typing errors to data verified against its source photo.

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 heats per day.
• Transcription time released for the melting supervisor at the end of every shift, which today is the last thing done and the first thing skipped.
• From 4-8 hours of latency between metal and data down to second zero, heat after heat.
• Errors eliminated in the one figure the whole traceability chain rests on: the heat number.

estimated payback 4 to 9 months depending on heats per day, combining the melting supervisor's transcription time released and the elimination of errors in the one figure the whole traceability chain rests on. *Estimate to validate*.

And the fair question from the production manager

“What if it misreads the magnesium percentage or the pouring temperature?” — reading fields from a form with a known layout is an anchored task, where the best models drop below 1.5% error [1]. And there is a second net: the extracted values appear on the line tablet next to the photo, each one checked against its physical range — a pouring temperature of 2,400 °C is rejected, not stored — and the melting supervisor confirms or corrects before anything enters the ERP.

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