Key findings

1. Where we left off in April

Our April paper covered the announcement itself: on 7 April 2026 the CMIP7 scenario team published its design paper in Geoscientific Model Development and, with it, retired the three reference scenarios that carried most of the climate literature since 2010. RCP8.5, SSP5-8.5 and SSP3-7.0 were labelled implausible. The word came from the scenario builders themselves, not from sceptics. If you have not read that piece, it covers where SSP5-8.5 came from, how a coal heavy modelling artefact became "business as usual" in tens of thousands of papers, and why the 97 per cent consensus figure measures something much narrower than most people think.

I wrote in April that the retirement looked like a one off correction, a clean up done quietly in the hope nobody would notice. Looking at this month's numbers, that read was wrong. It is not a correction. It is a process, and it is still running.

What happened since April is simple to describe. The April scenarios were placeholders, sketched while the integrated assessment models (IAMs) were still running. The final versions have now come out the other end of those models, and the finals are dramatically lower again. The embargo on the underlying IAM runs holds until 1 September, so this analysis uses the publicly available figure data archive that accompanies the published paper. I verified every number I quote below against that archive's compiled values adn the sources listed with them.

2. Draft to final: the numbers

The table below is the core of it. Cumulative CO2 from fossil fuels and industry, 2025 to 2100, in gigatons. The draft column is what accompanied the April announcement (archive tag 1.0). The final column is the version the earth system models will actually run (tag 1.1.1).

Cumulative fossil fuel and industry CO2, trapezoidal integral 2025 to 2100. ScenarioMIP-CMIP7 archive tags 1.0 and 1.1.1 (Zenodo record 20713982), compiled July 2026. The archive itself calls the data indicative and directs users to IIASA database for definitive research use.

Two things in that table are worth a minute each. First, the top end came down hard. HIGH lost 911 Gt between draft and final, on top of the 3,031 Gt already gone between SSP5-8.5 and the draft. Stack the full sequence: 7,380 to 4,349 to 3,438. The worst case scenario of the actually running model round carries 47 per cent of the CO2 of the scenario it replaced. For calibration, SSP3-7.0, the other retired reference, carried 4,799 Gt. The new worst case sits comfortably below it, down near SSP4-6.0 territory (3,077 Gt), a scenario nobody ever called scary.

Second, and I find this one nearly as interesting, the bottom end came UP. Very Low went from 56 Gt in the draft to 423 Gt in the final. The draft's near instant global decarbonisation was apparently no more plausible than the draft's coal boom. Both tails of the distribution got pulled toward the middle once real models with real constraints did the work. That is what you would expect from an honest modelling process, and credit where due, it is what happened. It is also an admission that the placeholder range shown to the world in April was, at both ends, not plausible either.

Whilst we are on overall spans: from SSP5-8.5 to the final MEDIUM, the benchmark the authors anchor to current policy, the drop is 4,852 Gt, or 66 per cent. Two thirds of the emissions in the old reference scenario have simply evaporated from the official view of the future.

3. The ladder: where the new scenarios sit against the old ones

Numbers in isolation slide past people, so line them up. The old SSP generation, the one AR6 was built on, spanned these cumulative totals: SSP1-2.6 at 916 Gt, SSP2-4.5 at 2,625, SSP4-6.0 at 3,077, SSP3-7.0 at 4,799, SSP5-8.5 at 7,380.

Now place the new set on that ladder. The final MEDIUM, at 2,528 Gt, sits below SSP2-4.5. Read that again, because I had to. The scenario that CMIP7 describes as current policies continuing at current effort, no new ambition, no strengthening, emits less cumulative fossil CO2 this century than the pathway the last generation treated as its moderate mitigation case. To be fair to the architecture, the two are not the same kind of object: SSP2-4.5 was built to hit a forcing target inside the old framework, and MEDIUM is a current effort scenario from a new one, so this is two totals landing either side of a line, not identical pathways. But the line matters. RCP4.5 and its successors were routinely presented as what aggressive climate policy might buy us. On the scenario community's own final numbers, we are already below that emissions total on autopilot. Put directly: the new reference range undershoots the 4.5 scenarios that were so often presented, wrongly, as what climate policy success would look like.

The HIGH scenario, it needs its own paragraph, because it will be the most misused object in climate science for the next decade. HIGH assumes emissions grow as high as the modellers coudl plausibly push them, "resulting from deep political, technological, and structural deviation from current trends". In plain terms: politics, technology and economics all reverse course at once, governments tear up existing climate policy, and the world re carbonises. The CMIP7 authors write, and I quote them exactly because this sentence will be ignored: "Clearly, this scenario is not a 'business-as-usual' scenario nor the no-policy reference scenario for the other scenarios." That is the scenario builders telling the research community, in advance and in writing, not to do the thing they did with RCP8.5. I reckon the warning will be ignored by some anyway. We will keep a list.

4. What the official futures now look like in degrees

Emissions are the input. Here is the output side, and I want you to notice how crowded this table is between 2.2 and 3.0°C. These are the published 2100 warming estimates from the major assessment bodies, all November 2025 vintage or newer, next to the new CMIP7 anchors.

As published by each body; definitions differ and the figures are not strictly like for like. CMIP7 values are FaIR ensemble medians vs 1850 to 1900, with the 33 to 67 per cent range in brackets. COP30: decision FCCC/PA/CMA/2025/L.24, adopted 22 November 2025, paragraph 8. Compiled from the ScenarioMIP-CMIP7 figure data archive and each body's published report, July 2026.

The official conversation about the future of the climate now happens inside a window of roughly 2.2 to 3.0°C on the central estimates, with the deliberate policy failure experiment at 3.45. These are medians, so the honest reading keeps the spread in view: MEDIUM's own 33 to 67 per cent band runs 2.72 to 3.30, and the full uncertainty tails run wider still in both directions. Even so, the 4°C and 5°C futures that fronted fifteen years of documentaries, central bank stress tests, coastal hazard maps and school strikes are not in the table. Not because a sceptic removed them. Because the IPCC's own scenario pipeline no longer produces them.

Worth saying once, plainly: a formally adopted UN decision at COP30 in Belém already acknowledged the 2.3 to 2.5°C range for full pledge implementation. The retreat is not a blog claim. It has a document number.

5. The part that bothers me: less CO2 in, more warming out

Here is the finding I did not expect when I opened the workbook, and the one I most want members to sit with. Between April and July, HIGH lost 911 Gt of cumulative CO2, a fifth of its total. Its projected 2100 warming went from 3.43°C to 3.45°C. Up.... yes, up. MEDIUM lost 570 Gt, and its projected warming rose from 2.85 to 2.98°C. Across the whole set, the implied warming per 1,000 Gt of CO2 climbed from 0.41°C in the draft to 0.59°C in the final. For reference, the old AR6 marker scenarios imply 0.42. The regression fits are tight in all three cases (R² of about 0.99). The relationship itself moved.

Before anyone objects that the draft had four scenarios and the final has seven, so the fitted line was bound to move: we controlled that. Refit both versions on only the four categories they share (Very Low, Low, Medium, High) and the jump gets bigger, not smaller: 0.41 in the draft against 0.63 in the final, a rise of more than half, with R² at 0.985 and 0.997. The finding survives its strongest objection.

Now the honest caveats, because we do the caveats here even when they soften my own argument. A near straight line between cumulative CO2 and warming is not itself news; that is the TCRE relationship these emulators are built around, so tight fits are expected by construction. What matters is the gradient. These temperatures are also not from full earth system models yet; they are medians from FaIR, a simple climate emulator, and the draft and final used different calibration versions (v1.4.1 against v1.5.0). The scenarios also differ in things other than CO2: methane, aerosols, land use, all of which shift the temperature for a given CO2 total, and a steeper slope can partly reflect non CO2 forcing rising while CO2 falls. The earth system model runs through 2026 and 2027 will give the real answer. This is a preliminary number from the project's own archive, and I present it as exactly that.

But step back and look at the shape of it. Think about the incentives for a moment. The projected warming number is the asset. It anchors the assessments, the funding cases and the headlines, and every group with a stake in climate urgency needs it to stay high. When the emissions side of the ledger deflates, the pressure moves to the other knobs: sensitivity, aerosols, carbon cycle feedbacks, the other gases. And here is the final scenario set arriving with a fifth less CO2 in the worst case and a warming number that went up anyway, via a recalibrated emulator and adjusted non CO2 assumptions. Maybe every adjustment is scientifically justified. The carbon cycle and aerosol literature moves, calibrations update, fine. But when the input halves over fifteen years and the headline output barely moves, you are entitled to ask which number is driving which. I reckon the 2100 temperature is the managed quantity now, the thing the process is reluctant to let fall, and the emissions are just the story that gets rewritten underneath it. That is my read, flagged as opinion. The slope change is not opinion. It is in the archive for anyone to check.

6. What the old scenarios already cost people

This is the part that stops the whole thing being an academic curiosity, and it is why I keep writing these papers. Scenarios are not just model inputs. They get baked into planning law, insurance pricing, lending decisions and rates notices, and they stay baked in long after the science has moved on.

New Zealand just gave us the cleanest example on record; the account below comes form the NZ press coverage and the documentary Managed Out. In July 2026 the climate minister, Simon Watts, wrote to local councils telling them they should not automatically treat RCP8.5 and other retired scenarios as the central or most likely future in decisions affecting infrastructure, land use, property rights or rates, because doing so "results in unnecessary costs being imposed on ratepayers, businesses, and communities". This came after years of councils drawing hazard lines on property titles off the back of consultant reports built on RCP8.5. On the Kapiti Coast, residents watched house prices fall and insurance premiums climb because of hazard notations on their LIM reports, the NZ equivalent of a property search. One homeowner in Managed Out said it straight: "They're using RCP 8.5 which is the most extreme outcome. House prices are on the way down, insurance prices are sky-rocketing."

And the man who helped build the scenario framework? Detlef van Vuuren, lead author of both the original RCP work and the new CMIP7 paper, has said publicly of the business as usual label: "No, that was always a wrong way to describe it." Richard Moss warned publicly in 2014 against treating RCP8.5 as the reference case. The warning existed. The literature used it as the reference case anyway, thousands of papers deep, whilst the people who knew better mostly stayed quiet.

I started counting how many Australian coastal hazard assessments and council climate risk plans lean on RCP8.5 and stopped at page three of the search results. That audit is a paper for another day. But if you have paid a levy, a premium loading or a compliance cost that traces back to a worst case projection, the number behind it has now been cut by more than half by the people who wrote it. You are allowed to ask questions about that.

7. Who decides, and why you cannot see the data until September

Now the governance question, because these scenarios steer trillions in investment and regulation and it is worth asking who writes them. The design paper carries about 45 authors and the final figure dataset lists 46 creators, with the two big integrated assessment institutes, IIASA in Austria and PIK in Germany, supplying eight of them between then (I counted the Zenodo record myself). The lead author heads the same Dutch group that led the RCP work fifteen years ago. The paper describes review webinars and comment rounds run inside the research community, and that is the whole of the accountability structure: internal review, invited feedback, no external board, no public vote. The community build the entire framework around its own internal judgement, and for numbers that end up in planning law that is worth saying out loud.

The finals come with a new wrinkle. The full IAM scenario details are embargoed until 1 September 2026. The ScenarioMIP terms, quoted verbatim: "The data can be used only for scientific purposes aimed at the production of peer-reviewed publications. The posting of the scenario data and/or analyses based on the data on social media or other dissemination (in the public or the press) is thus strictly prohibited." For wind tunnel data on a new aircraft, fine. For the numbers that will anchor the next IPCC assessment and a decade of public policy, an embargo against public discussion is a strange instinct. Whatever the justification, it tells you how the community sees the public: as a downstream audience, not a stakeholder.

And the same month delivered a reminder of how this machine has worked historically. A ProPublica investigation reported that the 2004 Pacala and Socolow "stabilisation wedges" paper, the one that told the world we already had all the technology needed to solve climate change, was funded and shaped by BP without disclosure, down to BP's chief executive suggesting the wedges metaphor itself at a London meeting. Socolow told ProPublica the paper would likely never have existed without BP's funding; the researchers maintain BP had no control over the scientific content. That paper gathered over 4,500 citations and a starring role in An Inconvenient Truth. Yet again, the correction arrives twenty years after the influence, and long after the policy got built on top.

8. What this does not prove, and what it does

Now the discipline section, because a sceptic who only audits the other side is just an advocate with a different flag.

This retreat says nothing directly about climate sensitivity, attribution, or the physics of CO2. The CMIP7 authors frame the emission cuts as good news, fairly earned: policies enacted, renewables cheaper than anyone projected in 2010, coal growth that never showed up. Hausfather and Peters made the core point in Nature back in 2020: the business as usual story was misleading. If you accept every word of AR6 physics, the scenario retreat still stands, and that is precisely what makes it powerful. On the establishment's own terms, the catastrophe pathway is gone. My own view, argued in the 600 million year CO2 paper and not repeated here, is that the sensitivity side of the equation deserves the same audit the emissions side is finally getting. The slope finding in section 5 suggests that audit will be resisted.

Nor does a 2.5 or 3.0°C central future mean nothing changes or nothing costs anything. It means the question becomes the one engineers ask, adaptation at what price, instead of the one activists ask.

One aside before the finish, off the scenario topic, and I flag it as exactly that, an aside. With this El Niño peaking, we have been reading the normalized disaster loss literature. Across 1990 to 2024, in the Munich Re and EM-DAT records, La Niña years carry median global weather disaster losses around 270 billion US dollars against roughly 155 billion in El Niño years (2024 dollars). In the US normalised hurricane damage record (Weinkle and colleagues 2018, extended through the 2025 season), the La Niña to El Niño median damage ratio ran about 4 to 1 across 1925 to 1997 and above 7 to 1 since, though the recent sample holds only 7 El Niño years, so treat the strengthening as suggestive. A cycle that size and a slow trend can coexist, so on its own this refutes nothing. The point is narrower: the year to year damage record breathes with the Pacific, hard, and almost no disaster headline you read this summer will mention it.

9. What to watch from here

Four dates and one prediction. First, 1 September 2026, when the IAM embargo lifts and the scenario papers publish; we will check whether the assumptions survive daylight. Second, the earth system model runs already underway since their northern spring, with results flowing through 2027; they will replace the FaIR estimates in section 5 and settle whether the steeper slope was calibration or something else. Third, the AR7 drafting cycle, where we find out whether the assessment leads with the 2.2 to 3.0 window or finds a way to keep 8.5 era impact studies alive. Fourth, the slow grind of jurisdictions following New Zealand; watch which Australian council is first, and which is last.

The prediction, on the record so you can hold me to it: within twelve months of publication, a major media outlet or a peer reviewed impact study will present CMIP7 HIGH as the business as usual future, in direct contradiction of the sentence its own authors wrote. When it happens we will name it here. The scenario community says the right things now. The machine that turns worst cases into baselines has not been switched off, and it is still hungry.

The retreat is not finished. I expect it has further to go.

Related Reading on Dayboro.au

References