The mold has to be there on its date

The problem

A slip never announces itself at once.

Today the date is defended with a planning sheet and a morning meeting fed by perceptions. The four factors that destroy it — material that does not arrive, an engineering change that lands late, real shop progress diverging from plan, and one tryout loop too many — live in four different places and none of them talk to each other. Nobody is lying. There simply is no complete picture until it is too late for anything but overtime.

• Behind a mold's date sits a vehicle launch, and behind a delay sit penalties, paid expedites and a supplier note carried into the next award.
• The slip builds quietly over weeks and surfaces when there is no margin left for anything but overtime.
• The four factors that destroy the date — material that does not arrive, an engineering change that lands late, real progress diverging from plan, one tryout loop too many — live in four different places and none of them talk to each other.
• Nobody is lying. The date is defended with a planning sheet and a morning meeting fed by perceptions because there is no complete picture until it is too late.

How it fits the IRIS system

Agents over four rings — it is all of the above, orchestrated.

Four rings, each with its own case behind it:

No single ring solves the problem: each already exists today in some form and the mold still ships late. Above the four rings the Agents keep one answer permanently live — will this mold hit its date, and if not, why and since when — with the cause identified and the alternative proposed. The decision stays human; what changes is that it is taken weeks earlier.

• Ring 1 · Supply. Connect reads supplier alerts and receiving documents and projects the impact of each delay onto the affected mold's date, by name.
• Ring 2 · Real shop progress. Job travelers by photo, legacy machine panels by camera, bench decisions by voice, spindle hours by API and human sign-off on the tablet. Progress stops being an estimate: it is the sum of its real details.
• Ring 3 · Engineering change. When a new data level lands, iLEAN instantly cross-checks which details are machined to the previous revision, which can be saved, what steel must be repurchased and how far the date moves. JIDOKA AI halts what has not yet reached a machine before steel and hours are spent on something obsolete.
• Ring 4 · Validation. Edge classifies defects shot by shot to shorten the tryout loop and blocks an unsafe press closing. SMED AI speeds up mold changeover on the validation presses so an extra trial does not eat a calendar day. The Agents close the dossier as it happens. Above the four rings, the Agents keep one answer permanently live: will this mold hit its date, and if not, why and since when? With the cause identified and the alternative proposed. The decision stays human; what changes is that it is taken weeks earlier.

See the full IRIS architecture →

Before and after

The four rings, and what each one closes

Ring	Which cases it draws on	What it closes
1 · Supply	Outside alert (3) + steel receipt (7)	The delay projected onto the mold's date, by name
2 · Real progress	Traveler (1) + EDM panel (2) + headset (4) + spindle hours (8) + tablet (5)	Progress as the sum of real, signed details
3 · Engineering change	Outside alert (3) + JIDOKA AI	Details to stop, details to save, steel to reorder
4 · Validation	Tryout vision (9) + safe close (10) + mold book (11)	Fewer loops, no damaged mold, dossier closed as it happens
Above the rings	The Agents	Will this mold hit its date, and since when not
SMED AI on the tryout presses	Ring 4	An extra trial does not eat a calendar day

from discovering the slip in the final weeks of the project → an early signal with the cause identified. From arguing the date with the customer on rhetoric → discussing it with the shop's own evidence.

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 6-12 months against the first slip avoided, counting penalties and acceleration costs: overtime, weekends, expedited freight, rescheduled press.
• The slip is discovered as an early signal with its cause, not in the final weeks of the project.
• The date is discussed with the customer on the shop's own evidence, not on rhetoric.
• And the larger, harder-to-quantify value: the reputation for delivery at the next award.

estimated payback 6-12 months against the first slip avoided, counting penalties and acceleration costs (overtime, weekends, expedited freight, rescheduled press). The larger and harder-to-quantify value is the reputation for delivery at the next award. *Estimate to validate*.

And the fair question from the production manager

"Do we have to deploy all twelve cases to get this?" — no, and framing it that way would be the way never to start. The rings are built in layers and each case pays on its own from the first month. With the job traveler and the outside alert you already have most of ring 2 and ring 1, which is where the early signal comes from. Every capture underneath is an anchored task, where the best models drop below 1.5% error [1], and every one is signed at the bench before it counts toward the date.

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