Required data is not proven data

Regulation (EU) 2024/1781, the ecodesign regulation known as the ESPR, starts from a list of product aspects in Article 5(1): durability, reusability, repairability, the presence of substances of concern, recycled content, recyclability, the recovery of materials, carbon and environmental footprint, and the waste a product is expected to generate. Article 7 turns those aspects into information duties and sets a floor while doing it: information requirements must, as a minimum, cover the digital product passport itself and the tracking of substances of concern across the product's life cycle. Around that floor sit repairability and durability scores and a carbon footprint, instructions on installing, using, maintaining and repairing the product and on returning it at end of life, and separate information written for treatment facilities on disassembly, reuse, refurbishment, recycling and disposal. Annex III then lists the elements the passport itself may carry, from the unique product identifier and commodity codes to the operator and facility identifiers behind them. Article 9(1) states the standard all of it has to meet: the data in the passport shall be accurate, complete and up to date.

The battery passport is the most explicit case, and the nearest one. Under Article 77(1) of Regulation (EU) 2023/1542, from 18 February 2027 every light means of transport battery, every industrial battery above 2 kWh and every electric vehicle battery placed on the market needs an electronic record. Annex XIII sets out what that record holds: the material composition, including the chemistry, the hazardous substances and the critical raw materials present; the carbon footprint; responsible sourcing as reported in the battery due diligence policy report; the recycled content; and, for the individual battery rather than the model, its state of health and a status field that reads original, repurposed, re-used, remanufactured or waste. That last field is the tell. The passport is not a snapshot taken at the factory gate. It is a record that keeps growing after the sale, and it is meant to still be readable when the battery starts its second life.

Two things happened this summer that make those elements travel. On 15 July 2026, six harmonised European standards for the digital product passport were cited in the Official Journal, covering unique identifiers, data carriers, data exchange, storage and persistence, application programming interfaces and system interoperability. On 20 July 2026, the EU's central Digital Product Passport Registry went live. A third clock runs alongside them: under the EU Deforestation Regulation, evidence duties for coffee, cocoa, soy, cattle, wood and rubber apply from 30 December 2026. The carrier architecture itself is deliberately left open. The ESPR requires open standards, an interoperable format and no vendor lock-in, and the Asset Administration Shell and the UN Transparency Protocol are two candidate architectures among others, not the legal format.

None of this means a company has to map everything today. What is actually owed is set product category by product category in delegated acts, and Article 8 leaves it to those acts to define the group, the requirements and the format. Much of what they ask for is aggregated rather than itemised: a score, a class, a footprint figure, not a line for every input. And access is tiered. The battery passport already splits into information open to the public, information open only to persons with a legitimate interest, and information reserved to notified bodies, market surveillance authorities and the Commission, while the ESPR builds the same graded access rights into Articles 10 and 11 and protects confidential business information alongside them. Today's obligation is narrower than the direction of travel. The direction is what this article is about.

A regulation can list an element and attach a date to it. It cannot make the element true. Every value that ends up on a passport was born somewhere else: on paperwork issued by another party, in another jurisdiction, under another legal system, often in another language and often years before anyone thought about passports. By the time it reaches the company that owes the passport, it has passed through several hands, each of which copied, summarised or re-keyed it.

An invoice states a quantity and a supplier, and is worth exactly what the party that issued it is worth. A land registry export states a boundary, and may be current, superseded, or simply about a different plot than the one that was harvested. A recycling certificate states a percentage, and that percentage rests on a chain of measurements no downstream reader ever sees.

This is where a single verdict over a whole document set collapses. A yes or no over a folder tells the reader nothing about which line inside it is solid and which one is an assumption, and the question that eventually arrives is never about the folder. It is about one attribute: this plot, this percentage, this date. A judgement that cannot be taken apart cannot be defended.

The answer is not a better verdict. It is a smaller one. Confidence has to be carried per attribute rather than per document set: deterministic checks score every attribute, and every attribute carries its source document and a confidence level on a hash-chained audit trail. One weak line then stays one weak line, visible as itself, instead of dragging a whole shipment down with it.

Automation belongs where the evidence is strong, and a named human belongs where it is not. Extraction from structured, signed documents is deterministic, and AI models never decide facts. Where confidence is low, the attribute is not published: a registered human attester resolves it first, reviews at attribute level, and that review is recorded on the audit trail. Review is confidence-driven, not blanket, which is what keeps it affordable. AI for scale, humans for the final word.

The last two requirements are about the reader rather than the writer. There has to be an integrity trail running from the source document to the published value, so that anyone asking where a number came from gets a document rather than an assurance. And verification has to work for a third party without calling the issuer: a passport issued as a signed credential, structured on open standards, that a buyer, a customs officer or an auditor can check independently. A public demonstration passport, clearly marked as a demonstration, shows the product scope, the attributes, the confidence levels and the status exactly as a real one would.

The confidence levels that decide what is published and what waits for a human are defined openly on the hosting page, not kept as a claim. And the public demonstration passport can be verified independently today, by anyone, without asking us for anything.

Every new delegated act adds a line to the passport. Our work is making each line worth trusting.