Until early 2026, a kilogram of American organic cotton leaving the gin gate was described in the sector’s reference data at 0.08 kilograms of CO2 equivalent. From March 2026 the same product is described at 1.77. Just over twenty-two times the figure. What moved was the reference evidence and the method behind it. Nothing in that revision says the crop got any worse.
Textile Exchange published a new cotton life-cycle assessment in March 2026: one kilogram of lint at the gin gate, cradle to gin, with soil organic carbon modelled separately as an indicative scenario and deliberately kept out of the headline result. As the figures were integrated by Carbonfact, the specialist that rebuilt its factor library on the new data, Indian organic cotton fell from 4.82 to 1.46. Turkish organic cotton rose from 0.64 to 0.95.
One revision. Three geographies. No single direction.
The old American figure was not conservative and it was not generous. It had stopped being the best available representation of the thing it described, and that was not visible from outside the dataset until farm-level information arrived to replace the estimate. Improved evidence moved the American figure up. In the same release, on the same day, it moved the Indian figure down.
Now take those numbers out of the technical report and put them where they are going. A column on a sourcing screen, beside a supplier name, next to price and lead time and an on-time delivery score. Two mills. Two product carbon footprints. One of them lower.
The eye does the rest.
The case that is not there
There is a fear in this industry that has never needed a citation, because everyone in it recognises the shape. Measure loosely and the number stays soft. Measure properly, which means metering the boilers, timestamping the batches, sending someone to stand in the dyehouse for a fortnight and record what is actually consumed rather than what the factor assumes, and the number becomes specific. Specific numbers can be argued with. Averages cannot. So the supplier who spends money on knowing more becomes the supplier who is easier to criticise, while the one who never looked sits comfortably inside the range.
I went looking for that case. Trade press, academic literature, grey literature, across many phrasings: a supplier that lost business because its own better evidence produced a worse number. It did not turn up. Not named, not anonymised, not as a line inside somebody else’s case study.
That is a hard constraint on what this article can claim, and it constrains it in one direction only. Sourcing decisions are private. Tender weightings are proprietary. Volume moves without an announcement, and a mill that loses a share of a programme is rarely told which variable moved it. A firm that depends on a customer has limited reason to make a public argument about that customer’s methodology. So the absence of a documented case means the phenomenon cannot be described as demonstrated, or systematic, or widespread. It does not mean it has never happened in a room that publishes nothing.
Unknown in private practice, unproven in public evidence. I would rather leave that open than close it in either direction, because closing it in either direction is the same error twice.
One distinction has to be set down before the argument goes further, or this becomes a complaint. If better measurement reveals that a mill genuinely emits more than anyone believed, and the buyer acts on that, the system is working exactly as designed. No supplier is owed protection from an accurate account of itself. That is not the case under examination here. The case under examination is the other one: two producers whose real environmental performance is close, whose reported numbers are not, and where the distance between the two figures is partly or wholly a distance between the evidence behind them.
The supplier-side position also has to survive its own counter-evidence, and there is some. Go looking for evidence that suppliers avoid measurement because they fear the answer, and what comes back instead is duller and more expensive: meters, systems, staff, software, verification fees. Cost and capability are the barriers on the record, not dread. And the nearest empirical work runs the other way. Song and colleagues, writing in the Journal of Operations Management in 2024, reported that suppliers facing customers with high environmental disclosure showed lower emissions rather than a retreat from measurement. The relationship they observed ran toward engagement, not away from it.
Resolution, not direction
So the simple story fails, and what replaces it is more useful.
Better data does not have a direction. It has a resolution. Sherwin and colleagues flew nearly a million aerial site measurements across American oil and gas production, covering 52 per cent of onshore oil and 29 per cent of gas, and reported in Nature in 2024 a weighted regional loss rate of 2.43 per cent, roughly three times the national inventory estimate for the regions surveyed. The methane was there. The inventory had not been lying. It had been coarse, and coarseness has no preferred direction. It can flatter a producer as easily as it can condemn one.
Which is why the other half of the concession matters. Primary is not a synonym for correct. A supplier-specific number can be badly metered, allocated across products by a rule nobody has revisited in four years, drawn from a period that does not represent the production being sold, or truncated at a boundary that quietly excludes the dirtiest step. Secondary is not a synonym for dishonest either. A well-chosen database value with a known distribution can describe a process more faithfully than a primary figure assembled in a hurry to answer a customer questionnaire. Granularity is not accuracy. It is only granularity.
On expected value, measuring is probably the right decision, and the case for it is stronger than the sector’s anxiety suggests. Indian organic cotton fell by seventy per cent when the evidence improved. What the evidence does not support is treating measurement as predictably punitive. It does not support treating it as predictably safe either.
Figure 1 · Better Data Does Not Have a Direction
One revision moved three geographies three different ways.
Three rows compare earlier and revised cradle-to-gin organic cotton carbon values. The United States rises from 0.08 to 1.77 kilograms of carbon dioxide equivalent per kilogram of lint, India falls from 4.82 to 1.46, and Turkey rises from 0.64 to 0.95.
United States
Organic cotton
Earlier
0.08
Revised
1.77
Up sharply
India
Organic cotton
Earlier
4.82
Revised
1.46
Down sharply
Turkey
Organic cotton
Earlier
0.64
Revised
0.95
Up moderately
Unit: kg CO2e per kg lint · cradle to gin
March 2026 LCA · soil organic carbon kept outside the headline result
One data revision. Three geographies. Three very different movements.
But expected value is not the only variable a supplier weighs. The cost is certain, immediate and theirs. The direction of the result is not knowable in advance. And the third term, the one nobody can price, is what the commercial system will do with the fact that the evidence is now better.
The second number
Increasingly, the architecture around a product carbon footprint can carry more than the value itself.
It can carry a primary data share, stating how much of the result rests on data from the actual value chain rather than on models. It can carry a data quality rating, scored across technological, geographical and temporal representativeness. Around those sit an assurance status, a system boundary, a database and its version, a reporting period, an allocation method, a functional unit. Not every footprint arrives with all of that, and plenty arrive with none of it. But where those fields exist, they describe between them something the headline figure cannot describe about itself: how much of it was measured and how much was assumed.
Call it the second number. It is not a formal term and no framework I am aware of issues it as a single metric; it is shorthand for the account of how well the first number is known. The first number answers what the estimate is. The second answers what kind of estimate it is.
Figure 2 · The Two Numbers
The estimate and the account of how well it is known.
Illustrative comparison of two supplier product carbon footprints. Supplier A reports 8.7 kilograms of carbon dioxide equivalent with visible evidence fields. Supplier B reports a lower 6.4 value, but its evidence is lower-resolution and more uncertain. The figure does not select a winner.
Supplier A · product
First number · carbon value
Second number · evidence context visible
Supplier B · product
First number · carbon value
Second number · lower resolution
Lower is not automatically more comparable · no winner assigned
A carbon number tells you the estimate. The evidence fields tell you what kind of estimate it is.
That distinction is not decorative, and the technical literature is clear about why. Heijungs, writing in the International Journal of Life Cycle Assessment in 2021, set the problem out precisely: the question a comparison actually asks is the probability that a randomly chosen unit of A outperforms a randomly chosen unit of B, and that is not the same question as which point estimate is smaller. His argument is that comparative practice conflates the two routinely. An eight per cent gap between two footprints means one thing when both figures are tightly constrained by measured data, and something else entirely when one of them is a sector average dragging its own distribution behind it.
A point estimate can look precise long before it is certain. Ranking is easy when uncertainty is invisible.
What the standards already refuse to do
It would be convenient to argue that the carbon-accounting frameworks have not thought about this. They have thought about it more carefully than most of the systems that consume their output.
ISO’s life-cycle assessment standards do not permit an inventory study alone to carry a comparative assertion intended for public disclosure, and attach conditions to such comparisons covering functional equivalence, method, data quality and independent critical review by a panel of interested parties. The GHG Protocol’s Product Standard states outright that claims of environmental superiority or equivalence against a competing product are not supported by it, and that additional prescriptiveness on accounting methodology is needed before results can be used to compare one product with another. These are not warnings about arithmetic. They are warnings about what the arithmetic is allowed to mean.
The sharpest instance sits in the Apparel and Footwear PEFCR, version 3.1, finalised on 29 April 2025 and valid to the end of 2027. It is a methodological instrument endorsed through the European Commission’s Product Environmental Footprint governance process rather than a piece of enacted law. Its subject is not the carbon figure on its own but the single aggregated score assembled across impact categories, and about that score it says this outright: the PEFCR “does not allow the use of the single score for business to consumer communications, nor its use for comparison against the representative products.” Communication must instead relate to at least the four most relevant impact categories or to additional environmental information. The same document requires a company-specific dataset to reach a data quality rating of 1.6 or better, and the study as a whole to reach 2.5 or better, before its results can stand.
Read that as a design principle rather than as a rule about carbon. A methodology can prescribe in detail how the evidence is to be assembled and then set an explicit limit on the kinds of comparison its own output is allowed to support. The limit travels with the method. It does not have to explain itself to mean something.
Which raises the question of whether anything comparable governs the commercial use of a product carbon footprint, at the moment one supplier’s value is set beside another’s. I did not find a published restraint of that kind.
Transmission is not enforcement
The data-exchange layer has done its part. The chemical sector’s Together for Sustainability initiative, whose current product carbon footprint data model dates from September 2025, marks its three data-quality rating fields mandatory from 2027: built into the exchange tools by the start of that year, reported by suppliers by the end of it. The PACT technical specification, on which TfS and Catena-X both draw, asks for a primary data share under its methodology while leaving the field optional in the schema itself, so a footprint that omits it can still pass schema validation. These are related institutional lineages rather than independent authorities converging on the same answer, and they should be read as one family, but the family has built the plumbing. The second number can travel.
Read the 2027 mandate again from a supplier’s chair. Mandatory to calculate. Mandatory to report. Nothing in it says mandatory to act upon.
That gap is the whole subject. Descriptive metadata tells the next system what kind of number it has received. A commercial rule decides whether that difference changes qualification, ranking, price, volume or review. They are different acts, and this territory merges them constantly.
Figure 3 · Where the Second Number Can Disappear
Transmission is not the same thing as commercial visibility.
A five-stage flow from measurement through product carbon footprint data exchange, sustainability platform, supplier scorecard and procurement decision. The first two stages can carry the carbon value and evidence fields; question marks at later stages ask which fields remain visible and influence the decision.
01 · Source
Measurement
02 · Exchange
PCF / data exchange
03 · Handoff
Sustainability platform
Which fields remain visible?
04 · Handoff
Supplier scorecard
Which fields affect the score?
05 · Handoff
Procurement decision
What reaches the decision?
The question marks are deliberate: this figure asks what survives each handoff. It does not claim that every system discards the context.
The standards can carry evidentiary context. The commercial question is what remains visible at the decision point.
I could not find a published buyer scoring rule that assigns a weight to a data quality tier or an assurance level as a scored line item. The search covered public-sector procurement specifically, because that is the domain which publishes its rubrics. Quality-price splits, in quantity. Net-zero questionnaires, in quantity. Nothing in anything I read that puts a weight on how well the number is known.
That is a finding about the public record and nothing more. Proprietary supplier scorecards were not observable and internal buyer platforms were not audited. It is entirely possible that a sustainability team inside a large brand retains provenance in its own system even where the tender interface shows a single cell, and that the loss happens at the interface rather than in the intent. The gap is a design question, not a charge.
The commercial layer is not uniformly indifferent, either. Volkswagen’s S-Rating has been a mandatory selection criterion across the group since 2019, and a negative rating bars a direct supplier from contract award: sustainability data operating as a hard gate rather than a preference. That is one manufacturer, in another sector, and it gates on a sustainability rating rather than on the quality of a carbon number. It is still a demonstration that this layer can be built to bite when somebody decides it should.
And the distribution matters more than the grievance. EcoVadis, surveying around two thousand suppliers for its 2026 procurement barometer, found 30 per cent reporting that they supply no carbon data at all, and 21 per cent reporting detailed activity-level data covering Scope 1 and 2 and selected Scope 3 categories. Before a buying desk reaches the problem of comparing two well-evidenced numbers, it often has to solve the problem of having two numbers at all.
The decision, not the supplier
Everything so far has been framed from the supplier’s side, which is where I sit and where the question originated. It is also the weaker half of the argument.
The buyer who ranks two mills on point estimates of unequal evidentiary quality is not being unfair to a supplier. They are making a multi-year sourcing commitment on a difference they cannot size, and then carrying that same number into their own Scope 3 position, their transition plan, their supplier engagement targets and eventually a public statement they will be asked to defend. The exposure does not stay upstream. It compounds downstream, in the buyer’s own accounts.
Other high-stakes systems worked this out a long time ago, and none of them did it out of consideration for the party being assessed.
Clinical evidence appraisal holds the certainty of the evidence apart from the strength of the recommendation built on it, and grades the two separately. A strong apparent effect resting on low-certainty evidence does not mechanically become a weak recommendation; other considerations enter, and they are argued out in the open. What the architecture refuses to do is let the certainty vanish inside the recommendation. The estimate and the confidence in the estimate stay on the page as two things. Military intelligence grades the reliability of a source from A to F and the credibility of the information from 1 to 6, and refuses to collapse the two, so that identical content from a B4 source is handled differently from an A1. Neither architecture protects the source. Both protect the person who has to act.
And where the quality of carbon evidence has been given an explicit price, the price runs against the fear rather than with it. Under Commission Implementing Regulation (EU) 2025/2621, and the correction that replaced its annexes in July 2026, an importer declaring on CBAM default values pays a mark-up on them: ten per cent this year, rising to twenty per cent in 2027 and thirty per cent from 2028. An installation declared on defaults pays the marked-up figure however efficient it happens to be. One that supplies verified installation-specific data reports its own number, however inefficient. In the one place I could find where carbon evidence carries a legislated cost, the cost attaches to relying on the default rather than the verified installation-specific number.
CBAM is a customs instrument at one border covering a handful of commodity sectors. It does not reach a denim mill selling to a brand, and nothing about its design should be transplanted into apparel sourcing without a great deal more thought. But it settles the distinction that everything else in this territory keeps blurring. A data quality field travelling beside a number is information. A mark-up is an incentive. An invitation is not a price.
Figure 4 · Metadata vs Price
Information can travel without changing the commercial consequence.
Two-panel comparison. On the left, data quality rating, primary-data share and assurance status travel as descriptive information. On the right, the CBAM default-value mark-up rises from ten per cent in 2026 to twenty per cent in 2027 and thirty per cent from 2028, changing liability. The CBAM example is not an apparel sourcing rule.
Descriptive layer
Information travels
The next system knows what kind of number it received.
CBAM default-value example
Information changes liability
10%
2026
20%
2027
30%
2028
Default evidence carries a legislated mark-up.
CBAM is a customs instrument for specified commodity sectors · it is not an apparel sourcing rule
Knowing that data quality is weak is different from making weak evidence costly.
The reporting layer built something adjacent years ago. The Science Based Targets initiative requires a company to recalculate its base year when a methodological change moves total base-year emissions by five per cent or more. Whatever the drafters had in mind, the effect of that rule is to hold a significant change of method apart from a change of performance, so that the two cannot be read as the same event. Those target-accounting rules have a mechanism for that. I found no equivalent operating between a buyer and a supplier.
What the column cannot hold
The architecture that follows from all of this is not complicated, and it is not a formula.
Environmental performance is one axis. Evidentiary quality is another. A mill with a high footprint and strong evidence is a known problem, which is a manageable thing to buy from. A mill with a low apparent footprint and weak evidence is not a better supplier; it is an unknown, and a column sorted on the figure alone will return the unknown at the top, because an unknown with a small number attached sorts exactly like a certainty. Better evidence should never excuse poor performance. Poor evidence should never be allowed to sit in the same column as certainty and be read as certainty.
Figure 5 · Two-Axis Supplier Comparison
Performance and evidentiary quality must remain separate.
A two-by-two conceptual matrix separates environmental impact from evidence quality. It shows lower impact with strong evidence, higher impact with strong evidence, low apparent impact with weak evidence, and higher impact with weak evidence. The matrix does not assign an automatic winner.
Evidence quality · weak to strong
Known strength
Lower impact
Known problem
Higher impact
Unknown behind a small number
Low apparent impact
Poor result, weak resolution
Higher impact
Conceptual architecture · no automatic winner · no scoring formula
Better evidence should not excuse poor performance. Poor evidence should not masquerade as certainty.
None of this requires a buyer to prefer a higher-emitting supplier because its measurement is better. That would be an absurd instruction and no supplier should ask for it. It requires something smaller and harder: that a sourcing decision be able to tell the difference between a gap in performance and a gap in knowledge, and know which one it is looking at before it ranks.
Picture the last moment before a number becomes a comparison. A compliance officer at a mill, late, entering a carbon figure into a customer’s portal. Give that portal a single field for the value and nothing else, and the rest of the scene writes itself: everything that officer knows about where the figure came from, what it assumes, what it leaves out and how much of it was measured rather than modelled has nowhere to go. Portals differ, and some ask for more than a value. The question is what the number carries when nobody asks.
The first number is an estimate of what was emitted. The second is the account of how well that estimate is known. A supplier is not owed protection from its own emissions. But a decision that can see only the first number may have found the better mill, or may have found the more generous set of assumptions, and nothing in front of it tells the two apart. It ranks them anyway, and files the result under carbon strategy.
