Data quality

Every carbon number is a measurement times a factor.

Somewhere behind every footprint is a small, unglamorous table: so much impact per kilowatt-hour, per kilogram, per tonne-kilometre. Which row of which table got used, and why, decides more about the final number than almost anything else in the assessment, and it is the part nobody asks about until an auditor does.

The short version

An emission factor is the impact per unit of an activity (per kilowatt-hour of electricity, per kilogram of fibre, per tonne-kilometre of freight), and a footprint is built by multiplying activity data by the right factors. Green Story draws factors from a ranked hierarchy of four source tiers, records which tier, dataset and version each one came from, and carries that rank through to the published result rather than discarding it once the arithmetic is done.

The mechanism

Activity data times a factor.

Underneath the modelling, a footprint is arithmetic of a very ordinary kind. You establish how much of something happened (kilowatt-hours drawn, kilograms of fibre spun, tonne-kilometres of freight moved) and you multiply it by a factor that says what a unit of that activity costs the environment. Do that across every input and add it up.

The activity data is generally the part people worry about, and it is the easier half. Meters read, invoices issued, weights recorded: these are facts a facility already tracks for its own accounting. The factor is the part that carries the judgement, because a factor is not a measurement of your supply chain at all. It is somebody else’s measurement of something they hope is similar enough.

Which is why two honest assessments of the same garment can differ substantially without either being wrong, and why the interesting question about any footprint is not what the number is but which factors it leaned on and how well they fit.

The distinction

Whose chain the number was measured in.

Primary data is a value from your own chain for your own order: the dye house’s meter reading for the month your fabric ran, the mill’s fuel invoice, the production record. Secondary data is a published figure for a process like yours, drawn from a life-cycle database or a study.

Both are legitimate, and any real assessment uses both. Nobody meters the electricity grid of a country per order, and nobody should: background processes such as grid mix, freight and bulk chemistry are exactly what established databases exist for. The problem is never that secondary data was used. It is when secondary data is used for the parts that actually differentiate your product, and the result is presented as though it described your supply chain.

Green Story works to a declared order of preference and the rank travels with the result. Your own primary data first. Then our own datasets, built from collection at supplier facilities rather than licensed in: 6,000+ proprietary data points across 40+ unique fibres, gathered on visits in 35+ manufacturing countries. Then established third-party databases for background processes. Then peer-reviewed literature where a fibre or process is new enough that nothing else exists.

An analyst may only move down a tier when the tier above has been tried and cannot be obtained for that specific input: per input, not per product. A single footprint routinely draws on all four, which is exactly why the split is stated on the face of the result rather than summarised away.

  1. Tier 1 · Primary

    Measured in your own chain

    Meter readings, utility invoices, production records and mill documents from the facility that made your product, attached to the order that caused the work.

  2. Tier 2 · Proprietary

    Our own datasets, built in the field

    Where an input cannot be collected for your order: datasets we built and maintain from primary supplier visits. Closer to your chain than any published average.

  3. Tier 3 · Secondary

    Established third-party databases

    Recognised life-cycle databases for background processes nobody collects per order: grid electricity, freight, bulk chemistry. Recorded with the version used.

  4. Tier 4 · Literature

    Peer-reviewed work, cited on the factor

    For emerging fibres and processes with no dataset yet. The citation is held against the factor rather than in a footnote nobody reads.

Where it goes wrong

The same word can mean a dozen different processes.

The most common serious error in a textile footprint is not a missing input. It is a factor that fits the word on the invoice rather than the work that was done.

Dyeing is the standard illustration. It covers a dozen genuinely different processes with a several-fold spread of impact between them, and the invoice says dyeing for all of them. Match on the word and you can be out by a multiple on the single largest block inside the factory gate. Matching properly means the technology, the substrate and the depth of shade, the detail a generic tool has no field for.

Region is the other one, and it is bigger than most reduction plans. A kilowatt-hour is not a kilowatt-hour: the same spinning process on two different national grids produces materially different numbers, and a grid factor taken from the wrong country is an error that dwarfs the improvements a brand is working hard to achieve elsewhere. Energy, water and treatment factors have to match where the work actually happened, which is one more reason the chain has to be traced before the footprint is computed.

And when two of your suppliers report very different numbers for what looks like the same process, that is usually a real finding rather than an error. Technology, machine age, shade depth and grid mix genuinely move a figure several-fold. Where the difference cannot be explained by the process, it goes to evidence review and, if it still stands, to a site visit.

The honest answer

What happens to last year’s report when a factor moves.

Factors are revised. Databases publish new versions, methods are updated, a national grid decarbonises and the figure for electricity there changes. This is the system working, but it means a number you published in good faith can stop being the number the same method would now produce.

Handled badly, this is where quiet dishonesty creeps into environmental reporting. The temptation is either to recompute everything silently, so that history appears to have always agreed with the present, or to leave old figures standing and let two incomparable numbers circulate without comment. Both destroy the thing the reporting was for.

What we do instead: identify exactly which products and reports are affected and re-run those rather than rebuilding everything. Every result carries the method and dataset version it was computed under, so two figures are never silently incomparable. The change is restated where the figure was published, with the previous version retained.

A correction nobody can see is indistinguishable from an error nobody caught, which is the whole argument for versioning, and the reason a claim made in 2026 can still be explained in 2029.

  1. On every result

    The tier and the dataset

    Which of the four source tiers this input came from, which dataset, and the version of it that was applied.

  2. On every result

    The measured share

    How much of the figure came from primary data, on the face of the result rather than in an appendix.

  3. On every result

    The analyst

    Who chose the factor and signed for it. An auditor asking why this factor gets an answer rather than a shrug.

  4. When a factor moves

    Targeted re-runs

    Only the affected products and reports, identified rather than guessed at, so a revision is a correction and not a rebuild.

  5. When a factor moves

    Restatement in the open

    Published where the figure was published, with the previous version retained. Nothing is quietly overwritten.

One more thing

Your own factors stay yours.

Brands that have done serious work often hold verified factors of their own: a measured figure for a process at a facility they part-own, or one established through a supplier programme. Where you hold one, it is used in place of ours and labelled as yours, so an auditor reviewing the result can see whose number is whose rather than having to ask.

That labelling matters in both directions. It means your investment in measurement shows up as yours rather than disappearing into a vendor’s dataset, and it means nobody later mistakes one of your figures for an independent corroboration of it. Provenance is the whole point of this page, and it applies to our own numbers exactly as it does to everyone else’s.

The words, gathered

Everything this page defined.

In the order you met them, in case you came back for one of them rather than for the whole page.

Emission factor
The impact per unit of an activity: per kilowatt-hour, per kilogram, per tonne-kilometre. The multiplier that turns activity data into a footprint.
Activity data
How much of something happened: energy drawn, mass processed, distance moved. The measured half of the arithmetic.
Secondary data
A published figure for a process like yours, from a database or study. Legitimate for background processes, and a problem where it stands in for what differentiates your product.
Background process
A process nobody collects per order: grid electricity, freight, bulk chemistry. What life-cycle databases exist to cover.
Measured share
The proportion of a result that came from primary data. The single most useful quality signal on any footprint, and the one to watch over time.
Dataset version
Which release of a factor source was used. Without it, two figures computed years apart cannot be known to be comparable.
Restatement
Republishing a figure that has changed, with the previous version retained, rather than silently overwriting it.
If you only wanted one answer

The shorter answers live on the FAQ.

Each of these is answered in a paragraph on the FAQ, which is where those answers live. This page is the long version of the same subject.

What provenance buys, and what it does not

Recording where a factor came from does not make it certain. It makes the result defensible, which is a different and more useful property, because it survives being asked about. We will tell you when a result leaned on a database for its background energy rather than letting you assume every kilogram was metered, and we would rather report that a first pass on a new supplier is thinner than you hoped than have the conversation that follows an unsupported claim.

Bring us your own product

Watch the boundary get drawn on something you made.

Pick one garment and we will walk it end to end with you: the unit, the boundary, what we would need from your chain, and what the result would and would not let you say.