Which emission factor datasets can I legally redistribute in a product?

Every carbon product is built on somebody else’s numbers, and almost every team assumes those numbers are either “open” or “not open”. We read the terms of all 246 sources in our corpus, one at a time, and that binary does not survive contact with the evidence. Forty-two of them use no recognisable licence at all — just bespoke wording you have to read and interpret. This page is what we found, including the sources we hold and may not republish, and the reasons in the publishers’ own words.

Licence review is the least glamorous job in building a carbon product and the one most likely to be discovered late, by a lawyer, after the feature has shipped. It is not hard work. It is just work that has to be done once per source, re-done on every new edition, and recorded somewhere a future colleague can find.

We have done it 246 times for our own corpus. The results below are generated from that record rather than written by hand, so the tables cannot quietly drift away from what the API actually serves.

Every source, with both permissions

The table renders from the source registry when you load the page, one row per source that holds rows in the corpus today. Two columns answer the two questions people conflate: whether GreenCalculus may serve the values with attribution, and whether you may republish them onward. It starts filtered to the second; switch chips to see the rest. Families are derived from the publisher’s wording, which is printed beside each and governs. No factor values appear here.

The same rows, as a file you can join to your own data: on Kaggle and on HuggingFace, CC BY 4.0, licence metadata only — no factor values.

There is no standard licence for this data

This is the finding that reframes everything else. Across 246 sources there are thirteen distinct licence families, and the largest single group is “none of the above” — bespoke terms written by the publisher, in prose, with no standard instrument behind them.

By row count the picture is friendlier than by source count: 15,762 of 17,074 rows — 92% — come from sources that permit republication. The restrictive sources are mostly small: a standards table here, a research paper there. That asymmetry is worth knowing, because it means the licence problem is rarely about volume. It is about the specific rows you happen to need.

Four permissions people treat as one

“Can I use this?” is four separate questions, and a source can answer them differently.

Two real examples from our own corpus show why the distinction matters.

ecoinvent answers differently within a single source: the methodology documentation is open, the quantitative data is not. You may describe how it works and cite it; you may not ship the numbers.

The GHG Protocol Product Standard is published with attribution on terms close to CC BY 4.0 — and is still not redistributable by a paid product, because the non-commercial condition bites on the fourth question rather than the third. A team that checks only “is it CC BY?” gets this one wrong.

What you may republish

Seventy-seven sources, 15,762 rows. Nearly all of them require attribution, and the attribution line is a condition of the licence rather than a courtesy — if you republish these values, reproduce the credit for each source you actually use.

What you may not, and why

Sixty sources, 1,312 rows. We hold and serve these — you can look one up and cite it — but we may not republish them in bulk, and neither may you without your own agreement with the publisher. The reasons below are the publishers’ terms as we recorded them at review, not our paraphrase.

Five traps that catch working teams

The source we held back although we were allowed

One entry in our corpus is flagged CC BY 4.0 by the publisher’s own metadata, and we exclude it anyway.

An IEA dataset on AI and data-centre energy arrives labelled with an open licence. Our earlier licence audit of the IEA recorded a general prohibition on third-party publication of derived footprints. The label and the audit disagree, and we have not resolved it with the publisher yet.

Check a source yourself

Search for an activity below. The result names the publisher behind it, the licence recorded at review, and whether that licence permits republication. This is the live corpus, queried from your browser with no API key.

How we read the terms, and how this stays current

Each source was reviewed once when it was ingested, and is re-reviewed on each new edition, because terms change — a source cleared in 2025 can carry a non-commercial condition in 2026. Four things get recorded: the licence name, the publisher, whether republication is permitted, and the sentence that decides it.

That record travels with the data. Every API response carries the licence of every source it touched, so a downstream consumer never has to come back to this page to find out what they may do with a value they already hold. It also decides what we can publish elsewhere: our benchmark of how accurate AI is on emission factors draws its questions only from the republishable subset, so the answer key can ship alongside them.

The tables above are generated from that record rather than maintained by hand, which is the only way a page like this stays true. A hand-written licence table is accurate on the day it is published and quietly wrong within a year.

Frequently asked questions