Zero-gap traceability, waste to destination

The problem

A traceability break is never one failure; it is a chain of small ones.

A break in intake-to-destination traceability is the sector's biggest legal and reputational risk. Every incident threatens the waste-management authorization and the relationship with construction clients who depend on that certificate for their own compliance.

• A form typed late, a weighing matched to the wrong truck, a contaminant that slipped past the cabin, a hopper signed off as empty when it was not. None of them is dramatic on the day.
• Each link looks acceptable on its own. Together they break the line between the ton that came in and the aggregate or landfill it ended up in.
• That break is the sector's biggest legal and reputational risk: every incident threatens the waste-management authorization, which no plant can afford to lose.
• And it hits the construction clients who rely on your certificate for their own compliance — they lose their evidence when you lose yours.

How it fits the IRIS system

Four cross-checked rings — every previous case, orchestrated by Agents.

Four coordinated rings:

No ring is new on its own; what is new is that they check each other. A load whose intake, sorting, hopper and batch evidence stop agreeing is held before it leaves, not discovered at the next inspection. The certificate you issue then rests on four independent sources that agree.

• the intake-scale record (form, delivery note and weighing) with no manual entry;
• the sorting camera confirms the fraction actually separated;
• JIDOKA AI blocks a new batch's start-up unless the hopper is confirmed empty;
• every batch leaves with its evidence pack cross-checking the three prior rings. SMED AI speeds up changeovers between waste types so the plant's pace isn't penalized.

See the full IRIS architecture →

Before and after

The four rings, and the gap each one closes

Ring	Built from	Gap it closes
1 · Weighbridge	Intake form (1) + panel (2) + delivery note (7) + scale-ERP link (8)	Which waste, from which site, at which weight — no manual entry
2 · Sorting	Belt vision (9) + supervisor's headset (4)	The fraction actually separated, not the one declared
3 · Hopper	Changeover validation (10), JIDOKA AI	No new batch starts on carryover
4 · Case file	Evidence pack (11)	Every batch leaves with proof of rings 1-3
Pace	SMED AI on waste-type changeovers	Traceability without slowing the plant
Gate	Two-tap validation (5)	No ring writes to the record without a person

Traceability rebuilt after the fact with gap risk → continuous, verified-in-the-moment traceability, by case file.

Impact estimate

Impact estimate — set against one serious traceability incident.

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, set against the cost of a single serious traceability incident — a figure to validate with management.
• Protection of the waste-management authorization, which is the plant's license to operate and the asset everything else depends on.
• Protection of the relationship with construction clients whose own compliance hangs on your certificate, and who choose their waste manager accordingly.
• And traceability moves from rebuilt after the fact, with gap risk, to verified in the moment, case file by case file, from the truck at the gate to the aggregate sold or the landfill cell.

Protects the waste-management authorization and the relationship with construction clients. Estimated payback 6-12 months against the cost of a single serious traceability incident. *Estimate to validate* with management.

And the fair question from the production manager

“Do we need all twelve cases before this works?” — no. The rings are built in layers and each case pays on its own; with the intake form and the weighbridge panel you already have most of ring 1. Each ring rests on anchored tasks — reading a form, a display, an image against its reference — where the best models drop below 1.5% error [1], and JIDOKA AI stops only on a verifiable discrepancy between sources. When the doubt is the model's, a person decides.

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