No liter should lose its identity between the tank and the customer

The problem

A liter loses its identity one hand-off at a time.

Between bulk receiving and delivery at the branch there are five or six pairs of hands and as many pieces of paper. Every hand-off is a chance for identity to break — product, batch, hazard, quantity — and a chance for a few liters to be left behind with no record. The first face is the one that keeps operations awake: a mis-identified drum is safety, claims and credentials at risk. The second is the one finance sees every month: bulk shrinkage left in the bottom of a tank, in a hose section, in the container that spilled and was replaced without a note, or in the manual inventory adjustment that squares the system without explaining anything. At month end, the difference between what came in and what went out is a number somebody has to justify without data. In a business with a tight gross margin, that number is not an accounting detail: it is the result.

• Between the tanker and the branch there are five or six pairs of hands and as many pieces of paper. Every hand-off is a chance for product, batch, hazard or quantity to break, and a chance for a few liters to be left behind without a record.
• The first face keeps operations awake: a mis-identified drum is safety, claims, a possible recall and the credentials that let you sell to large customers. Each control on its own passes; the concatenation is what escapes.
• The second face is what finance sees every month: bulk shrinkage in the bottom of a tank, in a hose section, in the drum that spilled and was replaced without a note, squared by a manual inventory adjustment that explains nothing.
• In a business with a tight gross margin, the difference between liters in and liters out is not an accounting detail: it is the result, and today somebody justifies it without data.

How it fits the IRIS system

Four rings that cross-check each other — it is all of the above, orchestrated.

No single ring solves the problem: each one exists today in some form and the drum still leaves wrong. The strength is in the cross-check, because what gives the broken identity away is four independent sources ceasing to agree, and what gives the shrinkage away is a mass balance that closes per batch instead of per month.

• Ring 1 · Source identity (cases 1 and 7): photographed receiving gives the liter its identity — product, supplier batch, certificate, weighbridge figure — and the blending order defines what it will become. Nothing starts without those two pieces matched.
• Ring 2 · The line and the label (cases 5 and 8): before the valve opens, the operator validates on the tablet the product-tank-container-label match; the labeler prints what the ERP publishes, pictograms included, with no keying in between; the scale returns the real weight of every container.
• Ring 3 · The eyes and the stop (cases 9 and 10): the exit camera reads the freshly printed label and checks it against the active order; the changeover cameras guarantee the line was clean. If anything fails to match, JIDOKA AI stops the line before wrapping.
• Ring 4 · The balance and the dossier (cases 6 and 11): Agents closes the mass balance of every batch — liters into the tank against kilos packed plus declared residue — flags the deviation the same day and attaches the evidence. Every pallet leaves with its traceable pack.
• SMED AI speeds up product changeover so the four rings do not penalize line productivity.

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 · Source identity	Blending order (1) + tanker receiving (7)	Which supplier batch and which formula the liter comes from
2 · The line and the label	Transfer check (5) + labeler and scale (8)	Product, tank, container and GHS label match; real weight per container
3 · The eyes and the stop	Closure and label (9) + changeover (10)	The label read against the order; the line proven clean
4 · The balance and the dossier	Lab COA (6) + evidence pack (11)	Mass balance per batch and a traceable pack per pallet
Supporting cases	Recovery panel (2), alerts (3), earpiece (4)	The still's yield, the late tanker and the supervisor's memory feed the same record
SMED AI	Across the rings	Changeover speed, so the rings do not cost line productivity

from discrepancies explained at month end with an inventory adjustment to deviations flagged the same day and with a cause · from mis-identified containers caught by the customer to discrepancies caught on the line · from traceability rebuilt by hand to traceability published.

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, measured against the first quarter with the mass balance closing per batch.
• Substantial reduction of unexplained bulk shrinkage: every point recovered drops straight to the result, in a business where gross margin is tight.
• Virtual elimination of shipments with the wrong product, batch or hazard label, and with them the recall exposure and the shipment rejected in full.
• And traceability that is published instead of rebuilt: the customer audit of a batch is a link, not a week.

substantial reduction of unexplained bulk shrinkage — every point recovered drops straight to the result — and virtual elimination of shipments with the wrong identity. Estimated payback of 6 to 12 months against the first quarter with the mass balance closing. *Estimate to validate with operations and finance, against the history of bulk inventory adjustments.*

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 tanker receiving and the blending order you already have ring 1, and with the labeler and scale you have the weights the mass balance needs. Each layer's cross-check is an anchored task, where the best models drop below 1.5% error [1], and JIDOKA AI stops on a verifiable disagreement between sources, never on model uncertainty.

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