Skip to content

How we know, and what a blank means

An empty field on an ActivateDigital passport is an answer. It means we looked, we did not get to something we were willing to stand behind, and we would rather say so than write a number that looks right.

Sources as at
28 August 2026
Share
LinkedIn X Email
On this page

Four answers, not two

Most product data systems carry two states. There is a value or there is a blank, and the blank means nothing in particular. This page is about how we establish whether something is known. What a business may itself publish when a field is genuinely unknown is a different question, and it is answered on Stating what you do not know.

A passport built on two states cannot tell a reader the difference between a duty that was discharged, a duty that does not arise and a duty nobody has checked. The engine carries four answers instead.

  • Resolved

    A value was established from evidence we can point at.

  • Not established

    We looked and did not reach a value. This is the state that becomes an anonymous blank box almost everywhere else.

  • Not applicable

    The question does not arise for this product. It is a finished answer rather than a gap, and it is the one most systems have no way of expressing.

  • Conflicted

    Two sources we would each trust say different things, and neither has been thrown away.

Completeness runs on its own axis. A field can be complete, partial or empty independently of which of the four states it sits in, because a partial answer that is honest about being partial is worth more than a complete one that filled in the remainder. The economic operator contact field is permanently partial on every product we have run, because no public route corroborates an address, and saying so is more useful than a tick.

Every state a reader sees is written in plain words rather than in a code. A state that has to be looked up is decoration.

What a missing value is actually telling you

Absence is where this subject goes wrong most often, and it goes wrong in one direction. Somebody looks, finds nothing, then writes that there is nothing to find. Those are two different claims and usually only one of them is supportable.

So an absence is typed before it is written.

The five kinds of absence, and what each one licenses a page to say.
TypeWhat it meansWhat may be written
StructuralThe interface makes the lookup impossible rather than difficult.There is no route, followed by why.
Access blockedThe record exists. Reaching it anonymously or for free is not possible.It exists. Confirming it this way is not possible.
Licence blockedThe record exists. Reproducing it needs a licence.It exists. Reproducing it needs permission.
Not found by usWe searched a stated scope on a stated date.We looked here, in this scope, on this date and did not find it.
Not yet researchedNobody has looked. This is our gap.Nothing at all. It is an internal gap rather than a finding.

We do not know that no number exists. We know we have not found one.

Only the first type supports a sentence saying that something is not there to be found, and even then the sentence explains what makes it impossible rather than asserting it. The substances database reference is the one field in the passport that meets that bar, and the attribute page for it sets out the interface reasons in full.

The distinction is not ours alone. A regulator's statutory review of the supplier information duty on substances in articles records that where no information was received it is not possible to see the reason: it could be non-compliance, or the substance could simply not be there. The same review records what happened where suppliers did respond, and both halves belong to the same document. A regulator declining to convert silence into a finding is the strongest external statement of this we hold.

There is a second trap under the first. An absence measured by a process that could itself have failed to look is not a fact about the records. If nobody wrote the extractor for a field, then finding none of that field says something about the extractor. Wherever a coverage figure of zero appears in our material, the caveat travels in the same sentence as the figure.

All of that is about how the absence is typed. The separate question, which arrives the moment somebody has to fill the field in anyway, is what a business may itself write into it. That has its own answer, its own instruments and its own decision tree, and it is set out on stating what you do not know rather than here.

Two questions to ask of any piece of evidence

Typed absence answers what happens when there is nothing. The harder case is the one where there is something: a real document, competently produced, expensive, and it still does not do the job. Two questions separate those, and they are independent of each other.

What the evidence covers

Companies rarely come unstuck because a field was empty. They come unstuck because the evidence they held did not cover the claim as it was worded, in that market, for that variant, at that granularity, under that chain of custody model. Two research legs reached that conclusion separately here, and a third arrived by another route.

So applicability is typed as well, on six facets. They are facets rather than steps, because they fail one at a time and independently: a document can be perfect on five and useless on the sixth.

A claim with six scope facets drawn beside each other rather than in a sequence, because each one can fail on its own. TYPED APPLICABILITY The claim, as it is actually worded Does the evidence held cover this claim, here, for this variant, at this granularity SIX FACETS. EACH ONE FAILS ON ITS OWN Market or jurisdiction Where the product is sold, and what that market lets you say about it Variant set Which colourways, sizes and versions the document covers Batch range Which production runs are named on the face of it, and which are not Chain of custody model What the model behind a certificate lets you say about this physical item Time window The period the evidence speaks for, and whether the claim sits inside it Not applicable, determined An explicit finding that a facet does not arise, recorded as one
The applicability facets. Six scope dimensions of a claim, drawn beside each other because they fail independently rather than in order. The sixth is a determination in its own right: recording that a facet does not arise is a finding, and leaving it blank is not. This is the model, not the machinery. How the facets are weighted, ordered or filled is implementation and stays inside.

Nothing above is a scoring system. There is no weighting, no resolution order and no enrichment method here, because publishing those would be publishing the machinery rather than the model, and the model is the part a reader can use against us.

What form the evidence takes

The second question is cruder and it is answered by looking at the document. What kind of thing is it, and what object does that kind of thing attach to. A test report attaches to a specimen. A site audit attaches to a site. A declaration attaches to whoever signed it.

Seven forms evidence arrives in, each with the object it attaches to, under a ceiling marking the graded vocabulary the estate has not bought. ABOVE THE CEILING Assurance grades, and the governed vocabulary that defines them Set out in a conformity assessment standard that is behind a purchase and has not been bought BELOW IT, WHAT THE FORM ITSELF SHOWS A prescribed test method What a defined procedure measures on a specimen, when it is followed An accredited test report What one laboratory measured, on the specimen it was sent, on a date A third party certificate That a site or a system was audited. Not that your goods are covered A supplier declaration What a supplier states, and what it holds. An assertion with a name on it A modelled calculation What a model returns for the inputs it was given An internal record What this business wrote down about its own process Nothing established No object at all. A state, and it is written as one
The evidence form ladder, and its ceiling. Seven forms and the object each one attaches to. The order is by how directly the form attaches to a physical thing, and it is not a ranking of truth: a modelled figure from a published public method can be better evidence for a value than a declaration nobody stands behind. There are no scores here and no confidence percentages, because the graded vocabulary that would supply them sits in a conformity assessment standard behind a purchase we have not made, and drawing that ceiling is more honest than inventing the grades under it.

The third rung is where most folders come apart. Read the face of a scope certificate and two schemes independently say the same thing on it: that it is no proof that any goods delivered by its holder are certified. The document that proves the goods is the transaction certificate, which is a different piece of paper that most folders do not contain. That distinction, scheme by scheme with document codes and versions, belongs on the documents page and is not restated here.

When two sources disagree

Quietly picking a winner is the most common failure in product data, and it is the hardest one for a customer to detect afterwards, because the output looks exactly like an output that had no disagreement in it.

Where two direct sources disagree, both values are kept with their source stamps and the disagreement is shown rather than settled. Where two care labels disagree, that disagreement is kept too, rather than being smoothed away by quietly publishing the gentler of the two instructions.

The clearest example is a number we could have published and did not. Our own durability coefficient sometimes disagrees with the coefficient the French regulatory calculator divided by. Rather than publish their figure divided by our number, the adjusted result is withheld and the calculator's own undivided figure is published, named as a disagreement.

What our own build currently shows

Zero rows are conflicted on the current 34 variant test catalogue. That is a property of a catalogue with one source behind it rather than evidence that the mechanism is unnecessary. It changes the first time a supplier declaration meets a merchant's own record.

The seven points at which evidence dies

Evidence is not usually destroyed. It is mislaid at a join, and there are seven joins where we keep finding it. Set them out in order and they make a spine, and that spine is the shape of everything downstream of this page.

The evidence spine in four stages, with the seven points at which evidence dies marked under the stage each belongs to, and the return path from a change in the evidence back to the published field left empty. THE EVIDENCE SPINE, AND THE FOUR STAGES THIS FAMILY FOLLOWS How we know The method, and the state a value resolves to Where evidence comes from The document, and what object it attaches to What a value will support The claim as worded, read against what is held After you publish The field is live, and its evidence keeps moving WHERE EVIDENCE DIES The absent request log The audit record collapse The batch to product join Copy generated from nothing The format wall Publication with no back link Silent expiry WHERE THE RETURN PATH WOULD BE Nothing carries a change in the evidence back to the published field A certificate lapses. A standard is withdrawn. Two sources split. The field is told none of it.
The evidence spine. Four stages, seven points at which evidence is lost, and one step with nothing in it. Across the chain of custody scheme rules read for this estate, not one operates automatic invalidation propagation, so the return path is drawn as an absence rather than as a weak arrow. No quantity is shown and no stage is presented as more common than another.

Ranked by what it costs to recover, and that ranking is ours rather than a measured one, the seven run in this order.

The seven points at which evidence dies, most expensive to recover first, each under the stage of the spine it belongs to.
Where evidence diesStageWhat recovery costs
The audit record collapseWhere evidence comes fromA structured audit record becomes a PDF with a number printed on it. The structure is gone, and getting it back means going to the body that issued it or testing again.
The batch to product joinWhere evidence comes fromA certificate names a batch and nobody records which products that batch became. Reconstructing it after the goods have shipped means reading records that were never kept for the purpose.
Copy generated from nothingWhat a value will supportA claim is written with no evidence object behind it at all. There is nothing to recover, so the only moves are to withdraw the claim or evidence it from scratch.
Silent expiryAfter you publishA document stops being current and nothing announces it. Cheap to fix for one document, expensive across a catalogue, because the hard part is knowing which ones.
The format wallWhat a value will supportEvidence sits in a vault as documents while the passport wants structured fields. Recovery is manual, one field at a time, and it is work somebody does twice.
Publication with no back linkAfter you publishThe value is live and nothing points from it to what it rests on. Recovery means deriving provenance again for a number people already believe.
The absent request logHow we knowNobody wrote down what was asked, of whom, on what date. The cheapest of the seven to start doing and the only one whose past cannot be recovered at all.

The last row is why the request log matters more than it looks. Without it, an absence cannot be typed: there is no way to tell not found by us from not yet researched, and those license completely different sentences.

Where a value came from, and what we show of it

A value without provenance is not a fact, it is a rumour. Every value on a passport traces back to something that was retrieved: which source it came from, which document, when it was retrieved and which part of that document held it.

Source material is never edited

What a source returned is kept as it arrived. If a supplier declaration is ambiguous, the ambiguity survives into the record rather than being tidied on the way in, so the reason a value is what it is can still be reconstructed a year later.

A refusal is evidence

Where a check declines to answer, the refusal is recorded with its reason rather than discarded. That is what makes an empty box interpretable. Something was tried, it did not produce a value we would stand behind, and the reason is on the row.

What is published, and what is not

Enough to understand a result and to disagree with it: the rule and the reason behind every value, the state it resolved to, the sources under it and the date each one was last checked. Where a public method exists, as it does for the French environmental figures, we use the public one and say whose it is, so anybody can reproduce the number rather than take ours.

What stays private is the implementation. That is an ordinary commercial position rather than an interesting one, and it does not cost the reader anything, because none of it changes whether a value is right. The things that do change that are on the page.

The practical test for any of this is simple. If a value on a passport turned out to be wrong, could anybody establish where it came from and who said it. If the answer is no, the passport is a database with a nice front end.

Knowledge that decays, and what has not run yet

Regulatory knowledge ages, and in this subject the cycles are knowable in advance rather than being a surprise. The Combined Nomenclature is republished every year, so a passport citing last year's commodity code is wrong in a way nobody notices. The REACH Candidate List is revised roughly twice a year, usually in January and June, so a bare count of substances goes stale silently.

Two dates are kept apart in the model and they are never merged. Sources as at is a fact about the world. It is the oldest human verification date among the sources a page rests on rather than the newest, because a page is only as fresh as its weakest leg. Page last reviewed would be a fact about us: a person read the sources behind the page and stood behind what it says.

The second date is not on this page, or on any other page here, and the reason is worth stating rather than leaving to be noticed.

What is running, and what is not

No page in this estate carries a human verification date, because no human review has been recorded against any source behind it. Every row in the source register starts at review required and stays there until a person completes one. The checking cycle described below is specified and has not yet run, so every sentence about it here is a commitment about what will happen rather than a description of what does.

When it runs, automated checking will have one job, which is to notice that a source moved. It will not decide what the movement means, it will not rewrite a sentence and it will not advance a review date. Automation may only move a source to review due. Only a human review moves a source to current, and a monitor that advanced a verification date every time it found no change would be manufacturing confidence rather than measuring it.

The commitment that follows from that is narrow and it is the one to hold us to. Where a page's review date sits well behind the sources it depends on, the page will say so on its face rather than leaving a reader to compare two dates and work it out. Until the first human review is recorded, the honest reading of the absent date is that nobody has yet signed this page, and a register of stale verification dates would be worse than no register at all.

What other people publish

Ranked evidence and trust taxonomies are not unusual, and anybody claiming novelty here should read four public models first. One is the Battery Pass Consortium data model of November 2024. The other three are an international transparency protocol from a standards body, a cross industry specification for exchanging product carbon data, and a sectoral data space working alongside a chemical industry assessment scheme. All four grade where a value came from.

Only the first of those four is registered as a source here, with an address somebody has checked. The other three are described by what they do rather than named, because we do not name what we cannot cite, and pointing a reader at an address nobody here has verified is the habit this page exists to argue against. Registering them is research that has not been done yet.

What none of the four holds is the interesting half. A value that could not be established. The difference between a record that is unavailable and a record nobody checked. Two sources that disagree. Evidence that something is absent, as against no evidence either way. And no state at all for a value whose support has since been withdrawn.

That is where the difference sits. Not in having a hierarchy of evidence, which several people have, but in giving absence, applicability, conflict, staleness and resolution state somewhere to live. If any of the four begins to model those, this comparison stops being accurate and the page changes.

The rest of this family, along the same spine

Everything above is the method. Nine pages apply it, and they are arranged under the four stages of the spine rather than under the state names an answer ends up in, because a reader arrives holding a situation rather than a state.

How we know

This page, and only this page. It holds the four resolution states, the five kinds of absence, the six applicability facets and the form ladder with its ceiling. The nine below borrow that vocabulary rather than restating it.

Where evidence comes from

  • Getting evidence out of suppliers

    The one statutory route to a supplier, who it actually belongs to, and what asking dependably produces, which is a record of having asked. This page says silence has to be typed; that page is what to do about it.

  • What each document is worth

    The form ladder taken document by document, scheme by scheme, with the clock on each one. This page names the rungs; that page tells you which piece of paper you are holding and when it stops counting.

  • Chemical data, and the object it describes

    Why chemical data that is real and well governed still describes a site, a formulation or a shipment rather than the garment. This page states the applicability question in general; that page answers it for one whole class of evidence.

What a value will support

  • Claims and substantiation

    The distance between a filled field and a claim about a product, which is where almost all of the trouble in this subject lives. This page models applicability; that page works it against the law on claims.

  • When the lab and the supplier disagree

    The conflicted state taken into one instrument, with the tolerances that decide whether there is a conflict at all. This page names the state; that page runs the procedure and names what changes on the day a result is issued.

  • Stating what you do not know

    What a business may itself publish when a value cannot be established, with the worked patterns named instruments already contain and the counter case where not knowing stops the sale. This page is about how anything is known; that page is about what you may write when it is not.

After you publish

  • Owning a published field

    Who inside a business signs a field off, what the registry regulation asks of the operator, and what it contains no procedure for. This page states the method; that page states who carries it and what a sign-off consists of.

  • What a scan tells you

    What is actually measured once a code is printed, with populations attached and no single rate. This page sets the rule about figures and denominators; that page is the hardest test of it we have.

  • What a recycler can use

    What reaches a sorting line, which is the object and whatever survives on it rather than your product record. This page is about evidence a reader can check; that page is about evidence a machine at the end of the line never receives.

You might want to read next

Since you have read this, these may answer the questions that usually come next.

Sources

  • Battery Pass Consortium data model
    IndependentRelevant provisions reviewed

    The only one of the four public precedents this page positions against whose address is registered here, which is why it is the only one named. The other three are described by what they do. Registering their addresses is research rather than a build act, and until it is done, naming them would be the habit this page exists to argue against.

    View official source

  • ECHA Candidate List of substances of very high concern
    34th updateRegulator dataReviewed in fullChecked 21 August 2026

    Named as an example of a source that is revised on a known cycle, so a bare count goes stale silently.

    View official source

  • Combined Nomenclature, annual amending regulation to Annex I of (EEC) 2658/87
    In forceRelevant provisions reviewed

    The other worked example of scheduled decay. Republished annually, so last year's commodity code is wrong in a way nobody notices.

    View official source

  • Statutory review of the supplier information duty on substances in articles
    Regulator publicationRelevant provisions reviewed

    Carries the strongest authority the estate holds for the proposition that silence cannot be read, and it comes from a regulator rather than from anybody selling a remedy. Both halves of the finding travel together here as they do on the supplier page, and neither is restated as a rate, because the review is held at the level of its findings rather than its underlying data.

  • Chain of custody scheme rules, scope and transaction certificate specifications
    Scheme rulesRelevant provisions reviewed

    Behind three things on this page: what a scope certificate says on its own face about the goods it does not cover, the chain of custody model as one of the six applicability facets, and the finding that no scheme read here propagates an invalidation to anything downstream. The full per scheme treatment is on the documents page rather than here.

Worth sharing?

Help someone else make sense of product passports.

LinkedIn X Email