The equilibrium assumption

Most economic models used in climate and energy policy start from the same place: markets tend towards equilibrium, and the people in them behave as rational, representative actors.

That orthodoxy runs through the dominant tools, most notably integrated assessment models (IAMs) and the energy-system models used to plan national grids. They are good at one question: how much of a technology is cost-optimal at an assumed carbon price? They are poor at the questions that decide whether anything gets built.

To make the maths work, the many hidden aspects of scaling a clean technology are assumed away into cost curves and supply curves. Who has to finance the plant, who has to buy its output, who has to permit it, and what each of them needs before they move: all of it collapses into a single line. So do costs that never appear on a balance sheet: social costs borne by the communities that host projects, environmental costs in land, water, energy and materials, and governance costs of verifying, permitting and coordinating across many actors. In a mature, stable market that can be a reasonable shortcut. In a new market defined by complexity and deep uncertainty, it hides exactly what decision-makers need to see.^[1]

The problem is not that the models are wrong about the technology. It is that they leave out the people who decide whether it scales.

Four futures, or thousands

Britain’s energy networks show the same problem. The regulator, Ofgem, must fund huge network investment for net zero while the path and timing remain deeply uncertain. In a 2021 Oxford Martin School paper, former Ofgem networks partner Maxine Frerk sets out why the tools used so far fall short.^[5]

The standard toolkit pairs cost-benefit analysis on a central forecast with a handful of detailed scenarios, and picks between investments using a rule called least worst regret. Frerk identifies four ways this goes wrong.

• Scenarios miss what mattersDistribution companies tested their plans against every official scenario. None of them envisaged the surge in solar PV that then caused real problems for some networks.
• Scenarios aren’t independent of decisionsShetland windfarms could not win contracts for difference without a grid connection, and the connection was not justified without the contracts. Planning for low renewables can produce low renewables.
• Decision rules hide their assumptionsLeast worst regret works as a black box. It claims to need no probabilities yet implicitly treats every scenario as equally likely, and adding an option can flip the choice.
• Flexibility looks like wasteAnalysis on a central forecast ignores the value of waiting, learning and adapting. Investing gives up the option not to invest if things turn out differently.

Frerk’s alternative is a shift from “predict-then-act” to “monitor and adapt”. Explore uncertainty more creatively than a fixed scenario set allows. Develop options that keep choices open: delay, stage, invest to learn, expand later. Evaluate them with Robust Decision Making, used as a decision aid rather than a decision rule, combining analysis with deliberation. As uncertainty and trade-offs grow, stakeholders need to be part of that deliberation, with models framing the debate rather than settling it.

“It is better to be roughly right than precisely wrong.”
Maxine Frerk, Oxford Martin School (2021)

Her closing principles could serve any clean-technology market. Resilience matters, and spare capacity is not necessarily waste. Agility is valuable. Certainty may never come, and at some point you have to act. Models should provide insight, not answers.

Loops, not lines

Complexity economics treats the economy as a system that is always forming, never settled.^[1] Three of its properties matter most for scaling clean technology, and a supply curve can represent none of them.

• Non-linear feedbackSmall changes in one place can have outsized effects elsewhere, through loops that reinforce or balance each other.
• Path dependenceEarly choices shape which futures stay possible. Infrastructure built for one technology biases what comes next.
• EmergencePrice swings, market concentration and stalls arise from many decisions interacting, without anyone designing them.

Network effects add another layer. A shared pipeline or storage hub lowers costs for everyone who joins it. Competition for a scarce storage site, grid connection or feedstock raises costs for everyone who needs it. Which way a market tips depends on how actors respond to each other, which is exactly what equilibrium models leave out.

Agreement isn’t action

Better models are only half of the problem. In almost every room we work in, people already agree that the transition must happen. Agreement still doesn’t turn into action.

Renée Lertzman’s work names why. The dominant theory of change assumes people lack information or motivation: tell them more, raise the urgency, remove the barriers. But care is rarely the missing ingredient. Fear, guilt and urgency tend to freeze people rather than move them, and talk of “gaps” and “barriers” renames the puzzle without explaining it.^[7]

Instead, Lertzman points to three forces that shape how people respond to hard issues, usually beneath the surface. She calls them the three A’s: anxieties, ambivalence and aspirations.

Hard truths are taken in only when the conditions allow it. In her words, knowing “often has nowhere to go”.^[8] Lertzman calls the answer a container: a quality of holding that lets difficult information be received, felt and integrated rather than deflected. Her practices for building one shape how we run every convening.

• Connection before contentStart with real relationship, then bring the data.
• Slow down and leave roomPause after hard material so it can link to what people already know.
• Ask rather than tellEvocative questions over correction. Reflection comes before skilful action.
• Small groups, over timeIntegration is a relationship, not an event.

Otto Scharmer widens the lens from the individual to the collective. What grows above ground, he argues, depends on the soil beneath it: the “social soil” of relationships, trust and shared sensemaking from which institutions grow. When that soil is depleted, norms erode, social bonds fragment and people lose the capacity to converse and collaborate. In his words, “the system eats its own foundations.”^[20] Lertzman describes what happens inside people; Scharmer describes what happens between them. Both point to relational work as critical infrastructure.

Better tools

We build decision-support on complexity economics. At its core is an agent-based model: many individual actors, each with their own goals, constraints and decision rules, interacting year by year so that market-wide outcomes emerge rather than being assumed.

• Different actorsDevelopers, buyers, financiers, policymakers, standards bodies and communities, each drawn from realistic distributions rather than one average type.
• Actors that adaptAgents revise beliefs and strategies as outcomes and policy change: by learning from payoffs, updating beliefs, or copying successful peers.^[3]
• Real dataCalibrated to transacted, deal-level data, with uncertainty reported in full rather than hidden in a central estimate.
• Many futuresPolicy choices and their sequence are stress-tested across many plausible futures, in the spirit of Robust Decision Making, to find the ones that hold up.
• Made to be questionedOutputs are shown as interactive visuals that a minister, a developer and a resident can all interrogate.

The model is one way of knowing among several. It offers an outside view of a system; the people living and working in it bring knowledge no dataset holds. We treat communities as Scharmer puts it, as “the subjects, not the raw material” of the data economy, which is why our data is governed under an open charter.^[20]

The aim is not to forecast. It is to ask how specific actors will respond when a price, a mandate or a standard changes, and in what order the levers should be pulled.

Better rooms

The other half of our work is relational infrastructure: the institutions, networks and trust that decide whether capital and policy actually deliver. A growing body of research argues this, more than technology, is often the binding constraint on scale.^[9]

We draw on relational psychology and on Brave Space leadership. Rather than casting one leader as the person who fixes things, Brave Space leadership sees the job as creating conditions in which people connect, share responsibility and act together. It asks for presence, honesty, deep listening and suspended judgment, which is what a room needs when people’s requirements genuinely conflict.^[21]

How we convene

• A red threadOne clear purpose that ties every session together.
• Who is missingEveryone a decision affects is represented, not discussed in their absence.
• Beginnings and endingsThe first and last twenty minutes are designed with care: they decide who feels they belong.
• Less content, more interactionSmall groups and real challenges instead of panels and talking points.
• Sustained, not one-offFollow-up and integration built in, so what was learned has somewhere to go.^[10]

There are good precedents. Citizens’ assemblies in Ireland and Taiwan, and the Global Citizens’ Assembly for COP30, show deliberation can shape hard national choices. Denmark’s folk high schools, and the Highlander Folk School they inspired, where civil rights leaders including Rosa Parks trained, show what patient capacity-building can do.^[20] Germany’s Coal Commission brought government, industry, unions, mining regions, scientists and environmental groups together to plan a coal phase-out.^[9]

Communities at the start, not the end

Too often, community engagement is a box ticked once the important decisions have been made. We think that is one of the main reasons good projects stall.

Holly Jean Buck and Sara Nawaz trace how social impact assessment, the main tool for weighing a project’s effects on people, has drifted from its origins. It was meant to inform an open debate about which impacts are acceptable. In practice it is usually commissioned by the developer, often serves to legitimise a project rather than improve it, and brings communities in at the end, when only small tweaks are possible.^[9]

• Too lateCommunities are rarely involved in defining which impacts matter, or which mitigations are acceptable.
• Too hard to joinTechnical jargon, distant venues, short comment windows and no support for the cost of taking part.
• Too narrowA cost-benefit frame struggles with place attachment, cultural heritage and livelihoods that cannot simply be replaced.
• Too shortParticipation dries up once the assessment is signed off, though the risks continue for decades.

They also separate two kinds of social impact. Place-based impacts belong to a location: noise, traffic, water, a changed landscape. Pathway-based impacts belong to the direction of travel: for carbon removal, whether it supports deep emissions cuts or gives cover for continued fossil fuel use. Project consultations end up hosting national debates they were never designed for, which frustrates developers and campaigners alike. Separating the two lets each conversation happen in the right room.

Above all, they argue that impacts do not live in a technology. They depend on how it is deployed and in what context: who owns a project, how its benefits are shared, how strong local protections are.

Their answer to the charge that this slows things down is direct. Community opposition is one of the main reasons large renewable projects are delayed or cancelled. Involving people at the outset, not on the margins at the end, could support “a more just, but also a speedier, transition.”^[9]

New evidence puts numbers on this for carbon removal. Fritz and colleagues asked 10,852 people in Brazil, Malaysia, Saudi Arabia, Italy, Norway and the UK to rate imagined removal projects that differed in who ran them, whether people were consulted, who kept the profits and how much CO₂ they removed.^[22] Consultation had the strongest effect on support, in every country and for every technology tested. Sharing profits with local communities came next, and not-for-profit arrangements also helped.

Three further findings matter for how we work. Opposition is not fixed: people who would oppose a project near their home respond just as strongly to how it is run. Most people want a real say: in the UK, about 44% think it fairest for local communities to decide whether a project is permitted, and across the six countries only 2–9% think no involvement is fair. And the authors recommend engaging at regional level and at the planning stage, rather than project by project and led by developers.^[22]

What this means for how we work

• In the room from the startCommunities and civil society take part in every ClimateNiche convening, from framing the questions to interpreting the results.
• Place and pathway, kept distinctLocal impacts and national direction are discussed in forums designed for each.
• Built into the modelSocial licence is an outcome of ownership and benefit-sharing choices, tested through our social-deployment scenarios.
• Accessible and openWe meet in accessible places and speak plainly, and our models, data and findings are published openly wherever practicable.
• Ongoing, not one-offRelationships continue beyond any single project, so the spiral can turn again.

The model itself is co-designed in these rooms. Decision rules, plausible ranges and the outcomes that matter come from structured workshops with each group of actors, communities included. Participants choose which experiments to run, and test the conditions under which a strategy would fail them.

Both at once

Each half fails on its own. Together they create the conditions for behaviour change across an ecosystem.

The evidence on carbon removal supports our thesis. In the six-country study, a project that removes a lot of CO₂ could not make up for an unfair process, and a fair process could not make up for a project that removes little. People judged effectiveness and fairness separately and wanted both.^[22] A model that only optimises the technology, or a convening that only builds trust, will each fall short on its own.

It also shapes how we expect change to spread. Markets far from equilibrium rarely change by central plan. Drawing on the physicist Ilya Prigogine, Scharmer describes such systems shifting through small “islands of coherence”: local groups that find a working way forward, then connect with others.^[20] So we start with one market, UK biochar, where the people involved can build shared understanding and show what works. Each success leaves behind a tested model and a set of relationships the next market can build on.

Two megatons, one gigaton

Net-zero pathways need carbon dioxide removal at gigaton scale by mid-century. Today, novel removal delivers about two million tonnes a year.

That is direct air capture, BECCS, biochar, enhanced weathering and ocean methods combined: roughly 0.002 gigatons, under 0.1% of all removal, growing about 40% a year.^[17] Almost everything else comes from forests and soils, whose permanence and saturation limits cap how far they can carry us.

• Technology maturityOf the novel pathways, only BECCS shows an encouraging learning rate, about 8%. The UK’s innovation needs assessment puts £31–62bn (DACCS) and £55–76bn (BECCS) of cost savings to 2050 on whether innovation is high or low.^[11]
• Geographic clusteringEurope hosts 48% of all novel removal projects.^[17]
• Infrastructure bottlenecksEuropean projects are spread across the continent, but CO₂ storage is concentrated in the UK, Norway, Denmark and the Netherlands.

Where removal models go wrong

Existing frameworks treat carbon removal as a monolithic supply curve. In practice the market is defined by heterogeneous actors, feedback loops and institutional frictions. Four blind spots stand out.

Two kinds of complexity sit on top of these. Market-creation complexity: fragmented technology pathways, untested public acceptance, and first-mover dilemmas under high capital costs. Market-led complexity: unresolved questions of who pays, a market running ahead of regulation, and private data that leaves policymakers and funders in the dark.^[4] Removal will not scale on supply-side momentum alone. It needs credible, long-run demand, and that is a question of how actors respond to policy, in both the what and the when.

A model of the removal market

Three strands of existing research each supply part of what is needed. The DSK agent-based integrated assessment model shows that interacting, boundedly rational agents change climate-economic dynamics.^[2] MUSE, Imperial College’s open-source agent-based energy framework, models heterogeneous investors with imperfect foresight across 28 regions.^[13] ADAGE lets agents learn how to respond to a changing environment, answering the Lucas critique.^[3] None focuses on the removal market. We bring them together for it.

Each agent is sampled from distributions of these attributes rather than fixed to a type, and each adapts. A financier tightens its risk model after a default. A standards body raises the bar after a scandal. A sceptical community warms as a project shares its benefits locally. Social licence is modelled as an outcome of deployment choices, such as ownership and local benefit-sharing, not as a fixed property of a technology.^[9]

Calibration and validation

The model is calibrated in four steps. Uncertain parameters such as learning rates are estimated as full distributions using neural ratio estimation, a simulation-based Bayesian method. Agent decision thresholds are fitted to behaviour. Variance-based sensitivity analysis identifies which few parameters drive most of the variation in outcomes, which tells us where better data matters most. Then the model is validated in layers, starting where data is richest.

Scenarios

Four scenario families, each at three intensities, are stress-tested with each group of actors. Crossing them asks not only which interventions scale removal, but under which social arrangements that scale keeps public consent. To make thousands of runs tractable, we combine surrogate modelling with explainable AI.

Limits, and how we handle them

• Many settings, one resultDifferent parameter sets can reproduce the same outputs. We report full distributions and focus on the parameters that matter.
• Numbers crowding out justiceQuantitative models can overweight what they can count. Place-based and justice questions live in the deliberation and enter the model as context and constraints.
• A young marketWith about 2 Mt of novel removal, out-of-sample validation is hard. We validate by matching cause and effect, not by fitting a forecast.
• Computing costSurrogate models and explainable AI make it possible to explore the parameter space efficiently.

Order matters

A model like this must be used to support live decisions. The question facing a policymaker is rarely “what is the right carbon price?” It is closer to this: if removals enter a trading scheme now and the standard is tightened later, how will developers and financiers respond, and will that sequence build the supply the scheme depends on, or undermine it? Two decisions of exactly this kind are being made now.

In the EU, the Commission has proposed bringing permanent removals into the Emissions Trading System on a one-in, one-out basis: one allowance retired for each removal certificate bought. Its arithmetic assumes removal costs fall below about €200 a tonne by the mid-2030s.^[18] Today’s traded prices sit well above that.

The second risk is the greater one. Retiring allowances for tonnes that never arrive lets emissions continue on the strength of promised removals. This is mitigation deterrence, policymakers’ deepest concern about removal and civil society’s central question of legitimacy. Because our model represents policy, finance and social licence together, it can test whether a given design, and the order in which it is introduced, reinforces emissions cuts or crowds them out.

In the UK, the Carbon Management Innovation Challenge brings developers, investors, researchers and government together in policy sprints to design the market for removals. It tracks progress with cost models that deliberately hold energy prices and financing constant.^[12] That isolates whether the technology is getting cheaper. Our model asks the next question: will market, finance, policy and social conditions let those savings become real projects?

Both decisions are being taken now, and early choices shape everything that follows. Public opinion is at a similar point. Political views barely predict support for novel removal today, but the authors of the six-country study expect that to change once real projects make costs and trade-offs tangible. They also find that people prefer not-for-profit models and that trust in private companies is consistently low, which sits uneasily with business models that rely on private investment.^[22] The time to test how ownership, benefit-sharing and finance can fit together is now, before the rules are set and positions harden.

Who it is for

1. Government policy teams designing demand-pull and de-risking instruments: the Department for Energy Security and Net Zero first, then the European Commission.
2. Financiers pricing removal risk, especially those stuck in the missing middle because they can’t inspect how a cost curve was derived.
3. Developers and buyers testing whether a policy design or contract structure is survivable for them.

Standards bodies, registries and civil-society organisations shape the model’s design and use its outputs. The public-good core, meaning the code, published scenarios, reports and visuals, is openly licensed and funded by philanthropic and public research capital. A commercial layer of bespoke scenario runs and dashboards supports long-term sustainability, following the precedent of the UK Offshore Wind Accelerator.

Why UK biochar first

• The cost gap is clearestPublished biochar cost estimates sit far below the prices buyers actually pay. A model calibrated on real deals has the most to correct.
• It fits the UK’s plansBiochar doesn’t depend on CO₂ pipelines or geological storage. The UK Innovation Challenge aims for ten or more commercial-scale removal projects that don’t rely on CCS, each removing at least 20,000 tonnes a year.^[12]
• Its producers have choicesA biochar producer can sell into agriculture or metallurgy instead of carbon removal. That puts a floor under the price it will accept, a behaviour a supply curve misses and an agent-based model captures.^[19]
• It’s small enough to calibrateA national market of tens of developers can be modelled against real project and deal data before extending to other pathways and countries.
• It can become an island of coherenceA market small enough for every developer, buyer, lender, regulator and host community to know one another, and to show what working together makes possible.

A model calibrated to transacted, deal-level data rather than to the theoretical literature is designed to break this cycle.

Ask the model

We have built a first working version. It simulates the UK straw biochar removal market year by year from 2026 to 2035, with four of the six agent types (developers, buyers, lenders and government) each making their own decisions. Here are three questions it can already answer.

Each answer changes the conversation. Certainty of sale matters more to lenders than a higher price. With the right support, the sequence of policies changes outcomes as much as the choice of tool. The cheapest public pound per tonne is one lenders can count before a plant is built. None of this is visible on a supply curve.

What success looks like by 2030

We will judge our hypothesis by three tests, and UK biochar is the first place we will run them. If none of them holds by 2030, this work will not have justified its cost.

1. iAt least one government policy process has used this approach to test a sequencing decision, and changed that decision as a result.
2. iiThe model is calibrated on transacted, deal-level data contributed by market participants under clear governance, not on theoretical cost literature.
3. iiiThe code and core outputs remain openly available and are in active use by researchers and institutions beyond ClimateNiche.

As a Community Interest Company, our datasets, models, code and research are open and free wherever practicable. Any surplus goes back into that public-interest work. Biochar is the first application; the same approach applies to any clean technology whose scale-up depends on many actors deciding together.