Their Own Scientists Drew The Curve In 1982. It Was Right.
Nobody had scored the old projections, so three researchers did. Average skill 72 per cent, the best of them 99, and 0.20 °C per decade. The data and the code are in a public repository.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
POTSDAM, Germany — Somebody finally scored them.
That is the news, and it is not what most accounts say the news is. The internal memoranda had been read, quoted and argued about for years. What nobody had done was the boring arithmetic: take the projections, line them up against what the temperature actually did, and grade them the way a meteorologist grades a forecast.
Three researchers did it in 2023.
What the grading found
An average skill score of 72 per cent, plus or minus six. The best single projection scored 99.
Between 63 and 83 per cent of the projections were consistent with what was subsequently observed.
The average projected warming was 0.20 °C per decade, which is — within the uncertainty — the same number the independent academic and government models of the same period were producing.
What a skill score is, corrected
Our first draft called it a measure of accuracy. Professor Ansgar Beaulieu-Ndiaye read that and told us it was wrong, and he was right, and his correction is the most useful paragraph in this story.
A skill score asks whether a prediction beat a trivial alternative — usually, assuming tomorrow will be like today. You can be accurate and have no skill at all, if the thing you were predicting was never going to change much.
Skill is the part that is not free.
Seventy-two is therefore not a mark for effort. It is a statement that these projections were doing real work against a baseline anybody could have used, across decades.
We ran it ourselves
The assessment's authors put their data and their code in a public repository.
This desk downloaded both and re-ran the computation. Our numbers matched theirs.
That is the least surprising sentence in this story and the most important one. Then we varied the choices — different baselines, different intervals — to see whether the answer moved. It did not move much, which is what a robust result looks like, and everything we ran is listed with this story.
What this is not
It is not a discovery of hidden documents.
The memoranda were internal, which is not the same as concealed, and several of the underlying projections were published at the time in the open scientific literature under their authors' own names. Anybody could read them. Some people did.
The new thing in 2023 was arithmetic, not archaeology, and this desk thinks the distinction is worth the paragraph.
What our correspondent could not establish
Who read them, and what they did.
The assessment scores projections. It does not establish an internal chain of decision, and neither can this newspaper. We spent three weeks trying and have nothing.
A retired research manager from an entirely different industry — he asked us to say that twice, and we have — described the ordinary mechanism without being asked about any company in particular.
"Research goes up. Positions come down. They are different pipes and they do not have to meet. Nobody has to decide anything for an organisation to say a thing its own laboratory does not believe. That is not a defence. It is worse, because there is nobody to sack."
The scientist's own ending
Professor Beaulieu-Ndiaye's full statement runs at the foot of this page and it takes the story somewhere this desk did not expect.
The people who produced this work, he says, were not doing anything unusual. They were doing competent atmospheric physics with the tools of their period, and getting the same answers everybody else was getting, which is what competent physics looks like.
He asked our readers to notice how ordinary that is — because the story is nearly always told as though a secret was being kept.
The physics was not a secret. It was in journals.
What differed was what an institution said in public. That is a question about institutions rather than about science, and he said plainly that he is not qualified on it and neither are we.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
The interesting fact here is not that documents exist. It is that somebody finally scored them, so we did what this desk does when a result matters: we downloaded the data and the code and ran it ourselves, then varied the choices to see whether the answer moved. It did not move much. Everything we ran is listed with this story and anybody with a laptop can repeat it.
Who we spoke to
- The 2023 assessment, Peer-reviewed; published in a general science journal. Read in full, including the supplementary material 2026
- Its public data and code repository, Open. Downloaded; our own data desk re-ran the skill-score computation May 2026 Our numbers matched theirs. That is the least surprising sentence in this story and the most important one.
- The internal memoranda and their graphs, Archival; reproduced in the assessment and in the repository. Read; the graphs redrawn for our own engraving 2026 Several of the underlying projections were also published in the open scientific literature at the time, under the authors' own names.
- Prof. Ansgar Beaulieu-Ndiaye, Climate scientist; not an author of the assessment and has not collaborated with them. Interviewed by video call twice May 2026 Explained what a skill score is and was firm that it is a grading method rather than a compliment.
- A retired industrial research manager, Named in our notes; worked in a different sector. Interviewed at length about how internal research reaches, or does not reach, an external position June 2026 Not connected to this company or this industry. He asked us to say so twice and we have.
Documents
- PX-2155 — Internal projection graphs, 1977–2003, filed with the 2023 scoring assessment and its public data and code repository accepted
What we could not confirm
- Who read the memoranda, and what they did about them. The assessment scores projections; it does not establish an internal chain of decision, and this newspaper cannot either.
- Whether the divergence was policy or incoherence. Large organisations routinely hold contradictory positions without anybody deciding to. Which this was is not established by anything we have read and we are not going to guess.
- How the scoring would change with different choices. Skill scoring involves choices about baselines and intervals. The repository lets you change them; we changed several and the picture did not move much, which is what a robust result looks like.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
Signal & LedgerAssessment Published; Shares Unmoved
Covered it as a market event, which it was not, in three hundred words.
Had that we did not
The publication date and the journal, correctly, on the day.
Left out
The numbers.
-
The Hollow PostTHEY HID THE PROOF FOR FORTY YEARS
Reported the internal work as suppressed and concealed until uncovered.
Had that we did not
That the public position and the internal work were different. That is the whole finding and it is true.
Left out
That the documents were internal rather than hidden, that several were published at the time in the open literature, and that what was new in 2023 was not discovery but arithmetic.
-
They Buriedthis newspaperTheir Own Scientists Drew The Curve In 1982. It Was Right.
Reported the scoring, which is the actual new thing, and the fact that the code is downloadable.
Had that we did not
The skill scores, and the repository address.
Left out
Our headline says 'their own scientists' and invites the reader to convict an organisation of something the scientists did well. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Prof. Ansgar Beaulieu-Ndiaye, climate scientist read the full draft; no deadline. Asked to correct a thing our first draft got wrong about skill scores, and then to add something. Printed unedited.
Your draft described a skill score as a measure of how accurate a prediction was. That is not quite it and the difference matters.
A skill score asks whether a prediction did better than a trivial alternative — usually, better than assuming tomorrow is like today. You can be quite accurate and have no skill, if the thing you were predicting was not going to change much anyway. Skill is the part that is not free.
So a score of seventy-two is not a mark out of a hundred for effort. It is a statement that these projections were doing real work, against a baseline that was available to anybody, over decades.
Now the thing I wanted to add.
The scientists who produced this work were not doing anything unusual. They were doing competent atmospheric physics with the tools of their period, and their results agreed with the results everybody else was getting, which is what competent physics looks like. Some of them published in the open literature under their own names.
I would like your readers to notice how ordinary that is, because the story is usually told as though a secret was being kept. The physics was not a secret. It was in journals. What differed was what an institution said in public, and that is a question about institutions rather than about science, and I am not qualified on it and neither are you.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.