open-climate.aiDocs

Editions and updates

Numbers checked against the live data on

A library is served one edition at a time: every number you see comes from the edition its publisher released and we loaded. This page says how to tell which edition that is, how to read the Versions section on a library page, when a newer edition is looked for, and what changes for you when one lands.

Which edition you are reading#

Several places name the served edition, and they always name the same one: the library card on Libraries and At a glance at the top of a library page, the provenance line under a factor's value, and the citation you copy from a factor page.

An edition can be old and still be the one to use. DESNZ serves its 2026 edition, published on 11 June 2026; EXIOBASE serves 3.8.2, published in October 2021 and deliberately held there. Age is not staleness: a publisher that issues every few years is not behind.

The Versions section#

A library page carries a Versions section once the library has an edition on record, and every library these docs use as an example has one. DESNZ versions is the one read here: one row per edition, newest first, four columns, and a link at the end of each row.

Column What it holds
Version The publisher's own label for the edition, such as 2026 or 3.8.2
Published The publisher's own date for that edition, a dash when the publisher gives none
Loaded The day our load of it finished, a dash when it was never loaded
Status Served, Superseded or Announced

The link at the end of a row, Source, opens the publisher's own page for that edition, and is absent when the publisher offers no such page.

Published and Loaded usually sit far apart, and a wide gap is not a fault in the table. Loaded is the day our load of that edition finished, not a turnaround from the publisher's release: a library imported for the first time this year carries a recent Loaded date however old the edition it serves is. Across the libraries these docs use as examples, the gap between the two dates runs from 56 days to 4.8 years. An edition is fetched, standardised onto the shared model, audited, and only then served.

Served, Superseded and Announced#

  • Served is the edition everything on this site answers with today: factor search, the factor pages, the API, an export and the open data download.
  • Superseded is an edition that was loaded and that a newer one has replaced. Its numbers have left all of those surfaces. They are kept in the database, so they can still be read, but only by querying it directly.
  • Announced is an edition the publisher has released and we have registered without loading it. None of its numbers are here: no factor carries them, and its Loaded cell shows a dash. Announced is a plan, not data.

The libraries these docs use as examples each serve a single loaded edition today, so their Versions section shows one Served row and nothing else. Superseded and Announced rows are what you will meet after a refresh, not now.

How a newer edition is looked for#

Above the table sits a line reading "Checked for updates" with a day, a verdict and how many signals were watched. It comes from a sweep of the publishers we watch, reading the signals that would reveal a new edition: the version label in a publisher's data catalogue, the checksum of the file we hold, the edition names printed on a download page. A library watched in more than one place has more than one signal.

The day on the line is the day that library was last swept, and no interval between sweeps is promised. Read the day itself rather than assuming the check ran recently.

The sweep looks at the publisher, not at the values. It can tell you that something newer exists; it never tells you that the number you are using is right.

The verdict is one of four, in the words the page prints.

The line says What it means for you
the publisher still shows the edition we serve Nothing newer was found that day
the publisher has released something newer, which we are reviewing A newer edition exists and is not loaded, and what is served has not moved
we could not reach the publisher that day The check failed, not the data. The served edition is unaffected
the result needs a human read, which is under way The publisher gives no signal a script can judge, so a person reads it

When a library is watched through several signals and they disagree, the line reports the most demanding of them. A library never claims to be up to date while one of its signals says otherwise.

A library page with no such line has had no sweep recorded yet. That is a fact about the check, not about the edition.

Note. "Something newer, which we are reviewing" changes no number. Nothing is swapped in until the new edition is loaded and appears in the Versions section as the Served row.

How often the data changes#

How often a library changes is the publisher's business, not ours, and it varies more than you would expect. The dates in the Versions section are the evidence you have: the DESNZ edition served here is dated 11 June 2026 and the MITECO one 7 May 2026, both from this year, while the EXIOBASE one is dated 21 October 2021. Some publishers announce the next edition, many do not, and a few revise the same edition in place instead of numbering a new one, which is why the Loaded date can move while the Version does not.

open-climate.ai promises no turnaround between a publisher's release and ours, and no notice before the switch. Plan a yearly re-run around the publisher's rhythm rather than ours, and read the library's Versions section before you count on a date.

One library is frozen on purpose. EXIOBASE stays at the edition served here because its later versions are published under a licence that forbids commercial use. Its update check reports that the publisher has something newer, and it will keep reporting it: moving to one of those versions would be a licensing decision, not a refresh. Do not read that line as a lag that is about to close.

What a refresh changes for you#

  • The address survives, the value need not. A factor keeps its page address across editions, so a link you saved keeps working and starts showing the new edition's number for the same activity. A reviewer who opens a citation you filed therefore sees today's edition, not the one you used, which is why the citation writes down the edition and the dates: they are what lets the two be told apart.
  • The number can move, and so can the years it applies to, the boundary and the unit the publisher prints it on.
  • A factor the new edition drops stops being served. It leaves factor search and the API rather than staying behind at its old value. An absence is explicit here: nothing silently keeps answering with a number the publisher has withdrawn.
  • Files you already have do not change. A download or an export you took stays as it was. A new one gives the new edition. Keeping an export checkable says what to write down at the moment you take one, so you can tell later whether anything in it has moved.

If you may have to show an auditor the exact figure later, keep three things at the moment you use it: the citation from the factor page, which names the edition and the dates; the export or download file the number came from; and, if you work in SQL, the identifier of the value itself. Pin a factor value to a release sets out what to store and how to read it back once the edition has moved.

That read back is the only route to a superseded number, and it needs SQL. Keeping your own copy means you never need it.

Where a change is written down#

The changelog records changes to the shared model and to the served view, not each library's refresh. An edition moving is the library's own event: its Versions section dates it, and its Changelog section says what changed inside it.

Next#