Methodology
A learning experience where people practise climate-policy decisions and see how their choices interact. The design connects evidence, constrained choices and reflection, with the model explaining what happens next.
8country perspectives
27policy options to weigh
4dimensions of performance
01 / The learning experience
The central question is: how do you reduce emissions while keeping a country able to act? Each design choice gives learners something to try, observe and explain.
Learners choose policies, prepare for crises and negotiate commitments. Results explain changes in emissions, budget, trust and preparedness, giving the next decision a reason.
Practise explaining a causal chain, rather than recalling a fact.
Each round has a climate budget, and every card in force is charged against it again each round. A separate capacity limit prevents implementing every policy at once. Learners weigh ambition, timing and what they are willing to give up.
Practise making and defending choices under constraints.
Four score dimensions and round-by-round explanations make trade-offs visible. Reflection prompts and session exports support discussion, comparison and a revised strategy.
Connect the experience to a transferable explanation.
These are intended learning goals. Software checks establish that the rules behave as designed; demonstrating learning gains requires evaluation with learners.
02 / The simulation loop
Select a country and a policy portfolio. Country baselines and policy settings shape the starting conditions.
The emissions model applies policy effects over time. FaIR uses the global pathway to estimate warming through 2100.
Round updates show changes in the economy and public trust. Crisis preparations and responses introduce further trade-offs.
Compare the four score dimensions, inspect what changed, and decide what to keep or revise.
A learner stages a coal phase-out and checks its cost, capacity demand and rollout time. Adding clean energy and grid investment changes the portfolio. After the round, the results make the emissions change and budget trade-off visible, giving the learner evidence to explain the choice.
Each round represents 8 years. Policies take years to decades to phase in; the long-term climate projection continues through 2100.
03 / Evidence and interpretation
Long-term warming is calculated with FaIR 2.2, a published climate model. Country baselines draw on emissions, energy and economic datasets. Policy cards reference IPCC mitigation evidence.
Policy costs, effect sizes, political responses and score weights include educational assumptions. The capacity limit (7 points per round by default; instructors can set 3–8) makes prioritisation tangible; it is a design rule, not a measured governance index.
Under the hood: a sector-based emissions model connects the policy portfolio to FaIR. A separate system-dynamics model tracks 27 physical and socioeconomic stocks, plus 4 tipping-point stocks; the policy cards' effects and prices are calibrated to it. The climate projection shows a 5–95% range from a 50-member ensemble, so learners can compare a central estimate with climate-response uncertainty.
50% weight
Emissions cuts, long-term warming and climate pressure.
20% weight
Budget, cost-effectiveness and growth.
15% weight
Public trust, preparedness and climate damage.
15% weight
Cooperation, agreements and institutional capacity.
The score is feedback on choices within the model, not a measure of a learner's real-world competence. Its weights express design priorities, and the temperature range covers climate-response uncertainty rather than every uncertainty in policy or society.
The model in detail
The rest of this page is the full reference: the climate model, system dynamics, policy assumptions, card prices, scoring, sources and limits.
04 / Architecture
Emissions are not read off pre-computed scenario curves. Each time a decision changes, a sector emissions model applies every country's policies to its 2024 fossil-CO₂ baseline, sector by sector: each policy phases in over time, and the cuts interact through the grid's carbon intensity, the extra load electrification puts on the grid, and policy synergies and conflicts. The global pathway, with the rest of the world added, is handed to FaIR 2.2, the current version of the simple climate model whose earlier v1.6 release was one of the emulators in IPCC AR6 WGI, which returns the temperature projection with an uncertainty range. Alongside it, at the end of each round a coupled system-dynamics model is integrated forward, with the same policies acting as lever parameters on energy capital, vehicle fleets, buildings, land, the budget, public support and institutions. Its tipping-point stocks add non-linear feedbacks as warming or deforestation approaches their thresholds, feeding carbon back into the next pathway or raising damages.
05 / Climate engine
Temperature trajectories are produced by the FaIR 2.2 simple climate model (Finite Amplitude Impulse Response; Leach et al., 2021). FaIR is an open-source, reduced-complexity climate model that maps annual emissions to atmospheric concentrations, effective radiative forcing, and global mean surface temperature via physically motivated impulse-response equations: a four-box carbon cycle and a three-layer energy-balance ocean. We use the fair-calibrate v1.4.1 parameter set (Smith et al., 2024), which is calibrated to CMIP6 models and then constrained with observed warming, ocean heat uptake and IPCC AR6 assessed ranges. An earlier release, FaIR v1.6, was one of the emulators used in IPCC AR6 WGI.
The model is spun up from 1750 with historical emissions of every species FaIR tracks (CO₂, CH₄, N₂O, aerosols and their precursors, ozone precursors, halocarbons and land use) plus solar and volcanic forcing, so 2020 temperatures and concentrations come out of the historical run rather than being set by hand. Emissions follow the calibration's “medium-extension” scenario through 2024. From 2025 the game's global fossil-CO₂ pathway replaces CO₂ from fossil fuels and industry; CH₄ and N₂O are scaled year by year by player decisions (sustainable agriculture, diet shift, and the mine methane a coal phase-out removes; the Global Methane Pledge commits a member to those cards and cuts nothing itself); land-use CO₂ follows the system-dynamics model's no-policy land path (3.8 Gt in 2025, 0.7 Gt in 2100) less two player inputs, the land-use emissions avoided by the land cards and the carbon Forest Restoration removes; and every other species, including aerosols, follows the scenario.
Whenever the round, the policies or the tipping feedback change, the browser builds the global emissions pathway from every country's sector baselines and the policies in force, adds a rest-of-world term for the countries outside the game (13.1 Gt CO₂ a year in 2025, the Global Carbon Budget 2025 world fossil total minus the eight playable countries, easing to 11.6 Gt in 2050 and rising to 17.5 Gt by 2100; from 2060 it is the system-dynamics model's no-policy world less the eight countries) and the carbon feedback from the tipping stocks, and sends it to FaIR in emissions-driven mode. FaIR runs a fixed 50-member subset of the 841-member constrained ensemble and returns the ensemble median with its 5th–95th percentile range, plus CO₂ concentration, total effective radiative forcing and cumulative CO₂ emitted since 2020. Temperatures are annual means relative to 1850–1900. The 1750–2024 history is the same for every player, so the server runs it once at start-up and each request resumes from the saved 2025 state, with results identical to a full run; a projection then takes about 0.4 s on the production server, with requests queued one at a time. The no-new-policies line is built the same way from the same countries, with no policies or tipping feedback: it reaches +2.85°C (+5.13°F) in 2100, with a 5–95% range of 2.14–3.91°C (3.85–7.04°F): the figure the system-dynamics model gives with no policy.
Each round covers 8 calendar years, so round 1 opens in 2025 and round 5 opens in 2057. A game runs 5 rounds unless the instructor sets 3 to 8. Policy changes made during a round enter the pathway FaIR receives gradually: each policy phases in along a smooth S-curve after any start-up delay. The ramp is calibrated to the system-dynamics model's response, from 1 year to 40 years (section 08 lists every card's). Cuts therefore build up across the whole game: a renewable, building or industry card adopted after round 1 is still short of its full cut when a 5-round game ends in 2065. The Policy Lab's rollout speed scales that whole timeline (rapid ×0.5, gradual ×1.5), and a pilot runs at 25% of the chosen level for one round before scaling up nationally. Neither changes the size of the final cut, only when it arrives.
Time horizon
2020 – 2100
Model version
FaIR 2.2, fair-calibrate v1.4.1 (50 members)
Output
Median and 5–95% range, °C (and °F) vs 1850–1900
06 / System dynamics
Beneath FaIR sits a coupled system-dynamics model whose state vector has 31 components (27 stocks for the climate, energy, land and economy, and 4 tipping stocks), integrated forward in time by scipy's solve_ivp with the LSODA solver; the tipping stocks are described in section 07. It describes one stylised world economy, normalised to 2020 (GDP = 1), and the selected player's policies for the round drive the server step. The browser fallback uses the game's policy records. The server integrates it once per round, starting round 1 from the model's own 2025 state; round steps are queued so each continues from the preceding server result. If a step fails or exceeds the 8-second timeout, the game keeps using its simplified browser model until reset; it does not resume a server history that has missed a period. The stocks span two classes.
Physical and technology stocks (21) are atmospheric carbon (C_atm), the carbon sink (C_sink), temperature anomaly (T_anom), clean, fossil, grid, storage and hydrogen capital (K_clean, K_fossil, K_grid, K_storage, K_hydrogen), coal plant's share of fossil capital (Coal_share, added in version 4.2.0 so that a coal phase-out retires coal only), fossil and electric vehicle fleet shares (V_fossil, V_electric), fossil and clean building heat (B_fossil, B_clean), forest and agricultural land (L_forest, L_agri), soil carbon (Soil_carbon), the industrial process-abatement share (I_clean), and four land stocks added in model version 3.1.0: restored forest (L_restored), farmland under improved practice (Ag_practice), the shift to low-meat, low-waste diets (Diet) and the no-policy forest-clearing rate (Clearing). In the game each round starts from FaIR's current temperature; in the validation runs FaIR's temperature and CO₂ replace the ODE's own climate stocks throughout. Every parameter is listed in docs/20-stock-parameters.csv (the file keeps the name it had when the model had 20 stocks) with its source and a confidence label. The carbon-sink rate is calibrated to Global Carbon Budget uptake and fitted to FaIR's CO₂, capital lifetimes come from IEA and IRENA, and baseline deforestation from FAO's Forest Resources Assessment, but 91 of the 161 parameters are labelled educational estimates.
Socioeconomic stocks (6) are unrepaired climate damage (D_damage), GDP, fiscal capacity (F), public support (S), policy capacity (P_capacity) and, new in version 4.0.0, output lost to unserved energy (Unserved_loss). Policy delivery runs through them: policy effectiveness is 0.30 + 0.70 × S × P_capacity (each capped at 0.85), so inside this model low support or weak institutions weaken every lever. In the game that gate does not reach the CO₂ cut on a policy card, which comes from the country emissions model (section 05). It scales how far policies move the indicator stocks and how much of a deal a partner delivers. These stocks are stylised, calibrated to make decisions feel realistic rather than to be macroeconomically precise. They still interact with the physical stocks in ways that reflect real political constraints: damage is 0.74% of GDP per °C² of warming and is repaired over about ten years, spending levers drain fiscal capacity, and public support rises as clean energy grows and, once emissions are falling, as warming is felt, but falls when spending weighs on the budget, when energy demand goes unserved, when carbon-price revenue is kept rather than returned to households, and when a phase-out closes plants with no support for the workers. With nothing pushing them, support and capacity rest at their 2020 values, 0.65 and 0.60, and gains taper as support nears consensus. Preparedness, diplomatic cooperation, energy affordability, employment pressure and inequality risk are tracked outside the ODE, in the game's own set of30 indicator stocks that the scores read; when the server's model answers, its results replace the matching index stocks (energy mix, grid, storage, EV share, building efficiency, forest cover, damage, public trust, governance capacity, fiscal capacity and carbon price).
Energy investment is demand-driven. Total clean plus fossil capacity follows energy demand, which grows with GDP at an elasticity of 0.58 at 2020 income that falls as income rises (efficiency decouples the two). Emissions from transport, buildings and industry follow income the same way, each with its own elasticity. Private investors, funded by the economy rather than the government's budget, replace retiring plants, build for the demand growth they expect and close any gap over about ten years. New capacity splits between clean and fossil by a cost logit (exponent 3): learning makes clean cheaper as wind and solar capacity grows (18% per doubling), the carbon price (lever × $250/t) makes fossil dearer, clean output the grid can't carry or store earns less, and public support tilts toward clean. A clean subsidy works like a tax credit, covering up to 50% of new clean capital cost. A phase-out blocks new fossil plants and retires up to 10% of fossil capacity a year, more slowly when nothing deliverable can replace it. Clean capacity is split into firm (hydro, nuclear, bioenergy, geothermal) and variable (wind, solar). A unit of variable capacity needs 1.11 units of grid, and storage that rises with the variable share of all capacity; the 2020 grid and storage are just adequate. A shortfall curtails wind and solar output, and fossil plants make it up until they run flat out; with no programme, grid operators work a backlog off over about ten years. Investors build ahead of a shortfall they expect, so no validation scenario leaves more than 0.1% of demand unserved in any year; demand left unserved costs output and keeps fossil plants open longer. Each spending lever has its own cost to the fiscal stock at full strength, from 0.03% of GDP a year for diet programmes to 0.47% for clean-energy subsidies (2.4% for all seventeen together), and six levers also carry a private cost for firms and households. The carbon price costs the budget nothing and raises revenue. The fiscal stock can spend at most 30% of itself a year; when it can't cover the portfolio, every spending lever is scaled back by the same funded share. Public grid and storage programmes close a share of the actual shortfall each year.
Policy levers enter the ODE as parameters held constant over each round. A Policy Lab card's level (10–100%) becomes a lever value of the same share, 0 to 1, through POLICY_LEVER_MAP (primary) and POLICY_LEVER_MAP_SECONDARY (secondary, with a scale factor); cards that share a lever add up, capped at 1. Each mapped card drives a lever through these mappings, with weights where needed (see section 05). Some cards bundle interventions that are deployed together in practice. Grid & Storage, for instance, drives both grid_investment and storage_investment at 0.6×. That mirrors how the two are built out together, and saves students from managing every sub-lever by hand.
Validation. Nine scenarios are run from 2020 to 2100 with FaIR coupled in, each against a benchmark set by its policy ambition rather than tuned to a target. These are scenario runs of the server model (version 4.2.0), not gameplay. With every lever at zero the model reaches 2.85°C (5.13°F) in 2100, inside the SSP2-4.5 range of 2.1–3.5°C (3.8–6.3°F) that it is benchmarked on: current-policies projections now sit there rather than in the high-end pathways (Hausfather & Peters, 2020). The two other no-new-policy scenarios, 2.76°C (4.97°F) and 2.84°C (5.11°F), are checked against the same range. Full Ambition, with every mitigation lever at full strength, ends at 1.94°C (3.49°F), inside the SSP1-2.6 range of 1.3–2.4°C (2.3–4.3°F). The five intermediate portfolios land between 2.39°C (4.30°F) and 2.57°C (4.63°F) and are checked for their place between the two. All nine pass. The rule was fixed before Full Ambition was first run; it is the third change of benchmarks and the first change of rule. The game's own no-new-policies line comes from the country emissions model, whose no-policy path is aligned with this model's, and also reaches 2.85°C (5.13°F).
14
Climate, energy and industry stocks
Carbon, temperature, energy capital and its coal share, fleets, buildings, industry abatement
7
Land stocks
Forest, farmland, soil carbon, restoration, farm practice, diets, clearing
6
Socioeconomic stocks
Damage, GDP, fiscal, support, capacity, unserved energy
4
Tipping point stocks
AMOC, WAIS, Amazon, Permafrost
07 / Non-linear dynamics
Four Earth-system tipping elements are implemented as additional ODE stocks, with thresholds adapted from Armstrong McKay et al. (2022, Science, 377:6611), a comprehensive assessment of tipping-element thresholds. Where the model's threshold differs from that paper's central estimate, the card below says so. Each tipping stock Tip_x ∈ [0, 1] evolves by first-order relaxation:
where σ is a logistic activation, k is a hand-set sharpness parameter (2–3 per °C, chosen to give a gradual onset, not fitted), and τ is a relaxation timescale, also hand-set: about 7 years for the AMOC, 12.5 for the Amazon, 20 for permafrost and 25 for West Antarctica. The pressure variable for each element is a function of the current temperature anomaly (and, for Amazon dieback, the deforestation fraction derived from L_forest). Stocks relax toward the current pressure, so they fall back slowly if warming drops; the hysteresis reported in the literature is not represented. Because the logistic is gentle, pressure builds well below the threshold: at 1.41°C (2.54°F), FaIR's warming for 2025, the WAIS pressure is already about 44%, while the AMOC pressure, with its 4.0°C (7.2°F) threshold, is still close to zero. As a stock activates it adds either an emission feedback (Amazon, Permafrost) or a damage multiplier (AMOC, WAIS). Once the server's model has run, every stock is reported in the Results dashboard under “Earth-system tipping elements”, which marks an element approaching from 10% and active from 20%. These two marks are labels on that screen; the feedback itself grows with the level and has no step at either. Amazon dieback and permafrost release add CO₂ in proportion to their activation, and each round's feedback is added to every future year of the next pathway sent to FaIR, so it steepens the temperature projection as it grows. AMOC and WAIS work differently: at full activation they add 12% and 3% to the annual climate damage the system-dynamics engine accumulates.
Separately from these four stocks, the game's tipping events (Greenland, West Antarctica, Amazon, boreal forest, AMOC, permafrost, coral reefs, Barents Sea ice) use the same paper's ranges rather than single lines. Each game draws a hidden threshold per element from its published range, with the central estimate as the median, and an element tips once FaIR's warming passes it, and stays tipped. The risk shown to players is the chance the threshold is already passed. The count of tipping events on the Results warming tile and in the Briefing is of elements that tipped in that round; it is a different figure from the four stocks above, which rise and fall with warming. A tipped element's effects then phase in over its published timescale — about 10 years for coral reefs, 50 for the AMOC, 2,000 for West Antarctica and 10,000 for Greenland — so committed ice-sheet loss barely registers before 2100.
Damage multiplier on annual D_damage accumulation. Reflects disruption of Northern European climate, intensified extreme weather, and altered monsoon patterns. The multiplier scales with the Tip_AMOC stock, adding up to 12% to annual damage at full activation.
Armstrong McKay et al. 2022, Table 1: AMOC collapse 4.0 °C (7.2 °F), range 1.4–8.0 °C (2.5–14.4 °F). Their Labrador–Irminger Sea convection collapse, at 1.8 °C (3.2 °F) with a range of 1.1–3.8 °C (2.0–6.8 °F), is an earlier, more regional tipping point the model does not represent separately.
Damage multiplier representing long-run sea-level commitment and coastal infrastructure loss. Although full WAIS collapse operates on centuries-long timescales, the damage signal reflects near-term intensified coastal hazard. Adds up to 3% to annual damage at full activation.
Armstrong McKay et al. 2022, Table 1: WAIS collapse 1.5 °C (2.7 °F), range 1.0–3.0 °C (1.8–5.4 °F), timescale ~2,000 years.
Emission feedback of up to +1.5 Gt CO₂/yr at full activation, representing the transition of the Amazon from a net carbon sink to a net source as forest dies and releases stored biomass carbon. Deforestation-driven pressure is derived from the L_forest stock, making it policy-responsive.
Armstrong McKay et al. 2022, Table 1: Amazon dieback 3.5 °C (6.3 °F), range 2.0–6.0 °C (3.6–10.8 °F); the 20% deforestation trigger is a model choice informed by the deforestation-threshold literature.
Gradual emission feedback from thawing permafrost releasing sequestered soil carbon. Real thaw releases CO₂ and CH₄; the model adds CO₂ only. Modelled as a linear pressure ramp from 1.5 °C (2.7 °F) to full pressure at 4.5 °C (8.1 °F), rather than a logistic, to reflect the gradual nature of high-latitude permafrost degradation. Up to +0.5 Gt CO₂/yr when fully active, and about +0.25 Gt CO₂/yr at 3 °C (5.4 °F).
Armstrong McKay et al. 2022, Table 1: boreal permafrost abrupt thaw 1.5 °C (2.7 °F), range 1.0–2.3 °C (1.8–4.1 °F); full collapse 4.0 °C (7.2 °F), range 3.0–6.0 °C (5.4–10.8 °F).
08 / Policy instruments
The Policy Lab offers 27 instruments across 7 sectors, and each card acts on the model in two places. In the emissions model it removes a fixed share of one or more sectors' CO₂ at full level, phased in after its lead time, and it can move the grid's fossil share as well, which cuts energy-sector CO₂ again. The coal phase-out retires coal power and nothing else: at full level, 95% of the country's own coal generation, over 35 years. Its level sets the pace of the phase-out, so half the level still retires about 63% of what full level does, at half the cost. Clean power replaces 55% of what is retired and gas the rest, so the solar, wind, nuclear, grid and carbon price cards still have that gas to displace. It cuts 64% of India's energy-sector CO₂, 40% of Germany's and 12% of Brazil's. Its cost is divided between the countries by the coal CO₂ each retires, about $3.9 a tonne at full level everywhere, so it is the one card that does not take the same share of the budget in every country: from 0.4% in Brazil to 24% in India. For Germany that is 88 Mt a year. Gas and oil power are not retired by any card: they are reduced only through the carbon price, clean supply and the grid. The CO₂ figure on a card is its cut for your country once fully phased in, with nothing else in force; it shrinks when other cards cut the same emissions, and its Sources panel lists each sector effect and the IPCC AR6 WGIII chapter that assesses the policy type. In the system-dynamics model the card sets one or two lever parameters, as a primary mapping or as a secondary lever at a defined scale factor.
The size and timing of each cut are calibrated to the system-dynamics model. For each of its levers, the cards that drive it, adopted together at full level by all eight countries in round one, avoid over 2025–2100 the CO₂ that lever avoids in that model started in 2025, scaled to the share of emissions the eight countries cover, so the eight are not credited with the rest of the world's emissions. The cards' relative sizes and each country's sector structure are kept, so a card still does more where its sector is larger. Lead times are the game's own (none to five years). The ramp that follows is the one that best fits the lever's response, the shortest of those within 2% of the best fit. After its lead time, a calibrated card's cut builds up over its ramp: 1 year for Forest protection, Carbon tax and Emissions trading (ETS); 14 years for EV Mandate and ICE Vehicle Phase-out; 15 years for Sustainable agriculture; 17 years for Public Transit; 25 years for Diet shift; 35 years for Coal Phase Out; 37 years for Grid & storage; 38 years for Building efficiency codes; 40 years for Solar power, Wind power, Hydropower, Geothermal, Nuclear Energy, Heat pump subsidy, Green Hydrogen Economy, Industrial Process Electrification, Industrial Decarbonization Standards and Carbon capture & storage. Most cuts therefore arrive over decades. Three cards are not calibrated: forest restoration, whose size is a published figure for each country, and urban greening and the national resilience programme, which have no lever of their own. Three cards cut no CO₂ and are calibrated on an indicator stock instead: institution building on governance capacity, the just-transition programme on public trust beside a coal phase-out, and the disaster recovery fund on the damage stock. Each has one coefficient, fitted so that the game's stock responds from 2025 to 2065 as the system-dynamics model's does.
The primary/secondary mapping architecture matters for model honesty. For example, Building Efficiency Codes drives building_retrofit as its primary lever and heat_pump_subsidy at 0.5× as a secondary, reflecting that building codes in practice bundle both fabric improvement and electrification. Urban Greening drives both afforestation (primary) and adaptation_investment (secondary at 0.5×), reflecting its dual role as a carbon sequestration and urban heat island mitigation instrument.
Coal phase-out drives fossil_phaseout; four renewable mandates (solar, wind, hydro, geothermal) and nuclear expansion all drive clean_subsidy and thus the K_clean capital stock; and a Grid & Storage card drives both grid_investment (primary) and storage_investment (secondary, 0.6×). The four renewable cards stay separate so students can see how the energy sources differ, but all of them contribute additively to the same clean-energy capital accumulation in the model.
EV mandate and ICE vehicle phase-out both drive the ev_mandate lever (additive); public transit expansion drives transit_investment. 3 cards.
Building efficiency codes drives building_retrofit as primary and heat_pump_subsidy at 0.5× as secondary, reflecting that building codes in practice bundle both fabric improvement and electrification. The Heat pump subsidy card drives the other half of heat_pump_subsidy, so the two cards together are the lever. Both act on each country's direct building-sector share of fossil CO₂ (Climate Watch), and the heat pumps' cut is net of the electricity they draw from the grid. 2 cards.
Green hydrogen, carbon capture & storage, process electrification and industrial decarbonisation standards drive the server's industry_abatement lever at weights of 0.5, 0.6, 0.7 and 1.0 respectively. Carbon capture also drives hydrogen_ccs as a secondary: capture at power plants, which adds about a tenth of the card's cut and is most of its price. This builds the I_clean stock and reduces industrial process emissions. Their sector cuts and added grid load remain in the separate browser emissions model; industrial fuel combustion is not yet coupled to the server's clean-grid share. 4 cards.
Forest restoration drives afforestation; sustainable agriculture drives sustainable_agriculture which directly affects the Soil carbon stock via a soil_gain flow term, and in the emissions model cuts agricultural CO₂ and trims CH₄ and N₂O. Country totals are fossil CO₂, so Forest Restoration cuts no fossil CO₂. It removes land carbon instead: restored area times a sequestration rate, growing over 20 years and capped at each country's cost-effective potential (Roe et al. 2021), sent to FaIR as a land-use removal. It also raises the forest stock. Forest protection drives deforestation_protection and avoids a share of a country's net land-use emissions (nothing where land is a net sink). Diet shift drives diet_shift: mainly methane and nitrous oxide, with a smaller land-use share. Land shares of different cards add up. 4 cards.
Carbon pricing and the emissions trading system both drive the carbon_price lever additively, so deploying both compounds the price signal. Their Configure drawer sets where the revenue goes; returning it to households sends household_dividend. Institution building drives institution_building and the just-transition programme just_transition; neither cuts CO₂. Treaties joined on the Negotiate page cut nothing themselves (they commit a member to cards, section 10), and class proposals passed there bind members to their own cards and count in the Governance score; they move no stock by themselves (section 10). 4 cards.
National Resilience Programme drives adaptation_investment. The Disaster Recovery Fund drives disaster_recovery: it prevents no damage and repairs D_damage faster. Urban Greening drives afforestation (primary) and adaptation_investment (secondary at 0.5×). The resilience and greening cards mainly slow the rate at which D_damage accumulates, so the resilience card shows “Protection / up to X%” rather than a CO₂ figure. They also trim a little CO₂ in the emissions model: urban greening 3% of building emissions, the resilience programme 2% of building and 2% of agricultural emissions at full intensity. 3 cards.
09 / Policy cost model
The simulation does not attempt to forecast exact national policy costs. A card's price is one year of the public cost of the lever it drives in the system-dynamics model. A card in force is charged 8 years of that price every round it stays in force, at the level it is in force at; the figure on a Policy Lab card is the cost for one round. Raising a card charges the difference from that round on. Lowering or repealing one costs less at once, and its cut falls back at the rate it was built. If the cards in force cost more than a round's budget, the most recently adopted are lowered when the round ends. The status bar, the plan panel, the Briefing and Results all show the round's money and points with the cards in your plan counted, so staging a card moves them at once. Financing chosen in Configure (climate finance, carbon credits, a larger private share) takes at most 50% off a card's cost. The budget is a share of GDP too, the same in every country, so a card takes the same share of the budget everywhere, with one exception, the coal phase-out. What differs between countries is what the card buys: it cuts more where its sector is larger. Cost per tonne is a year's cost divided by a year's CO₂ cut once the card is fully phased in, for your country at the level shown: the cost for a round, as printed, divided by 8 and by the Mt printed beside it. A card's cut is what it adds beside your other cards, in force or in your plan. When other cards cut the same emissions the card's own cut shrinks and its cost per tonne rises: the card then says what it would cut alone and names the card that overlaps. From $1,000 a tonne the card and Configure flag it before you save.
A round's budget is 8 years of the country's annual climate budget, which is 0.44% of its GDP in every country: from $81B a round for Brazil to $1,092B for the United States. The budget refills every round, and money not spent does not carry over. The annual budget is the country's GDP times an assumed public share of global energy-transition investment: BloombergNEF's $2.1 trillion for 2024, times 25%, divided by 2025 world GDP ($136B for the United States, $86B for China and $22B for Germany a year). The 25% is the IEA's estimate of public finance across all energy investment, not clean energy specifically; applying it to the clean-transition total is a teaching assumption, and the 2024 investment figure is set against 2025 GDP on purpose, the latest of each. The result is a consistent allocation for the game, not an observed national climate budget.
Rapid rollout costs ×1.3 and gradual ×0.8; a pilot costs ×0.68 while it runs. A card adopted or raised after round 1 costs 3% more for each round that has passed, up to 25%, and up to 30% more while public trust is below 40. The rate is fixed when the card is adopted. No card can take more than 60% of the budget at full level. Only Grid & Storage, at 64%, reaches the cap, and all 27 cards together cost 340% of the budget in the United States (more where the coal phase-out costs more).
A country has 7 capacity points a round (a class can set 3 to 8). Adopting a card takes 1, 2 or 3 points, by how hard it is to pass, at any level up to 50%, and 1 more above 50%: a full-strength programme takes more to run. Raising a card in force takes 1 point; lowering or repealing one takes none. The Policy Lab does not stage a card the points left cannot take, and points reset every round. A card's level is a per cent of its full strength, set on a slider from 10% to 100% in steps of 10. Add stages a card at 50%, and an untouched card shows its figures at 50%; a card in your plan shows the level and rollout it is set to. Its cost rises in proportion to the level. Its cut need not: where it rises more slowly, Configure says so with the card's own figures. Every screen gives a level in per cent. A deal or a class proposal asks for a card at 10%, 30%, 50%, 70% and 90% or more.
Each lever of the system-dynamics model has a public cost at full strength, as a share of GDP a year. A card that drives the lever is charged that cost for every year it is in force, 8 years of it a round. The formula is:
To find the share, the levers a group of cards sends at full level are run in the model from 2025, and their public cost, averaged over 2025–2100, is split between the group's cards in proportion to the CO₂ each avoids alone. The coal phase-out comes to 0.036% of the eight countries' GDP a year, 8% of the budget for the eight together, and is divided between them by the coal CO₂ each retires: $2.2B for the United States (1.6% of its budget), $17B for China (19.3% of its budget), $0.3B for Germany (1.5% of its budget). The coal phase-out retires coal power and nothing else: at full level, 95% of the country's own coal generation, over 35 years. Its level sets the pace of the phase-out, so half the level still retires about 63% of what full level does, at half the cost. Clean power replaces 55% of what is retired and gas the rest, so the solar, wind, nuclear, grid and carbon price cards still have that gas to displace. It cuts 64% of India's energy-sector CO₂, 40% of Germany's and 12% of Brazil's. Its cost is divided between the countries by the coal CO₂ each retires, about $3.9 a tonne at full level everywhere, so it is the one card that does not take the same share of the budget in every country: from 0.4% in Brazil to 24% in India. Public transit takes 32% of the budget, carbon capture 20%, nuclear 15%, building efficiency 13% and forest restoration 13%, the three industry cards 5% together and sustainable agriculture 5%.
Exception: the EV mandate and the ICE phase-out drive a regulatory lever with no public cost. They are charged its private cost instead (3% and 7% of the budget), because the game has always charged a mandate. The cost parameters and their sources are listed, one row each, in docs/20-stock-parameters.csv.
The four renewable cards share one mandate, priced at the clean-energy subsidy lever's public cost. Grid & Storage is priced the same way from the grid and storage levers. The formula is:
The mandate is 48% of the budget in every country: $65B for the United States, $41B for China and $11B for Germany a year. Grid & Storage comes to $88B for the United States, $56B for China and $14B for Germany, which is 64% of the budget (after the 60% cap: $82B, $52B, $13B). Each renewable card (solar, wind, hydro, geothermal) is one technology's slice of the mandate: its share of the country's deployable renewable build-out (the same share that sizes its CO₂ cut; large hydro for Brazil, Canada and Russia, geothermal mainly for the United States and Japan), priced at that technology's cost of generation relative to the share-weighted mean (IRENA global average for new projects, 2024: solar $0.043, onshore wind $0.034, hydropower $0.057 and geothermal $0.060 per kWh). A player selecting all four pays the full mandate and gets the full build-out.
A carbon tax or an emissions trading system costs the budget nothing. The revenue the government keeps is added to the budget every round the card is in force: 0.120% of GDP a year for the carbon tax and 0.131% for the trading system at full level, 56% of a round's budget for the two together. Revenue starts when the price does: a carbon tax has a 3-year lead time, so the round it is adopted in credits 5 of its 8 years. Returned to households as a dividend (a setting on both cards), the revenue adds nothing to the budget, and public trust gains each round instead.
The rule behind the two shares: the carbon revenue the government keeps in the system-dynamics model (2.26% of GDP a year, averaged over 2025–2100), credited to the game budget in the proportion the game budget (0.44% of GDP) stands to that model's fiscal inflow (4% of GDP), and split between the two cards by what each avoids alone. Repealing the card ends the revenue from that round. With the dividend, the price no longer lowers energy affordability.
Like the other prices, this rests on the system-dynamics model's parameters, which are sourced where a published figure exists and labelled educational estimates where none does.
The adaptation cards are priced like the others, from the levers they drive. The formula is:
The National Resilience Programme drives the adaptation-investment lever; the Disaster Recovery Fund drives the disaster-recovery lever, which the resilience card carried a part of until the fund had a card of its own; Urban Greening drives afforestation and adaptation investment. The adaptation lever's cost, 0.25% of GDP a year at full strength, rests on published estimates of 0.1–0.5% of GDP (de Bruin, Dellink & Tol, 2009) and on UNEP (2023), which puts developing countries' needs alone at $215–387 B a year this decade. The resilience and greening cards cost $45B and $40B a year for the United States; $28B and $25B a year for China; $7.3B and $6.6B a year for Germany. That is 33% and 29% of the budget in every country; the recovery fund is 20%. A country's vulnerability score does not change the price.
| Policy | USA | China | Germany | Source |
|---|---|---|---|---|
| Coal phase-out | $2.2B | $17B | $0.3B | ETC, 2023 |
| Grid & storage | $88B ($82B) | $56B ($52B) | $14B ($13B) | ETC, 2023; IEA, 2023 |
| One renewable card | $5.3B–$28B | $1.2B–$20B | $0.3B–$5.6B | ETC, 2023; IEA, 2021; IRENA, 2025 |
| Carbon pricing (revenue, each card) | +$37B–$40B | +$23B–$25B | +$6.0B–$6.6B | System-dynamics model |
| Climate adaptation | $27B–$45B | $17B–$28B | $4.4B–$7.3B | de Bruin et al., 2009; UNEP, 2023 |
| Climate budget (a year; a round is 8) | $136B | $86B | $22B | BNEF; IEA; World Bank |
All values are in billion USD a year, at moderate speed; the Policy Lab lists the cost for a round, 8 times these. They are not national policy expenditure forecasts. Figures in brackets are what the budget is charged once the 60% cap applies; the renewable row spans the cheapest of the four cards to the dearest, and the adaptation row the Disaster Recovery Fund to National Resilience. Sources are those of the levers' cost parameters (ETC: Energy Transitions Commission, Financing the Transition). Country inputs: GDP from the World Bank and, for carbon-pricing revenue, 2024 fossil CO₂ from the Global Carbon Budget 2025 (via Our World in Data).
10 / Governance model
The governance layer has two interlocking components: a domestic capacity budget that limits how many reforms a country can legislate in one round, and an international negotiation system in which players make deals, join treaties and, in a class game, vote on class proposals. A deal or a passed proposal binds its members to their own cards; it cuts no emissions by itself.
A country has 7 capacity points a round (a class can set 3 to 8). Adopting a card takes 1, 2 or 3 points, by how hard it is to pass, at any level up to 50%, and 1 more above 50%: a full-strength programme takes more to run. Raising a card in force takes 1 point; lowering or repealing one takes none. The Policy Lab does not stage a card the points left cannot take, and points reset every round. One point is a narrow, low-ambition instrument, two a sector-wide one, three a transformative, cross-sectoral one. Players cannot adopt or change more than their remaining capacity allows.
The system is an educational simplification of a real and well-documented phenomenon: institutional bandwidth, political capital, legislative throughput, and administrative implementation capacity are genuine binding constraints on climate policy ambition that operate independently of available finance (IPCC AR6 WGIII, Ch. 13). The 7-point default is inspired by that framing but is not derived from a published governance index.
This forces a trade-off absent from single-player dial simulators: a transformative energy transition in one round means giving up other sectors, mirroring the political-economy reality that transformative reforms consume disproportionate political capital even when fiscal cost is manageable.
In a class game any country can put one proposal a round to the class on the Negotiate page. There are four kinds, and each works only through the members' own cards:
Each country has one vote and can change it until the instructor ends the round, when the votes are counted. A proposal passes if more classmate countries voted yes than no and at least two countries in all said yes. Abstentions and countries that did not vote count on neither side. It binds the countries that said yes and nobody else.
A passed pledge, floor or pool becomes a class agreement: a deal with one party per bound country, kept on the class server and settled by the same code as a two-country deal each time the instructor ends a round. A classmate's promise to hold a card is checked every round against the cards that country submitted; a round below the level is a broken promise. A pool's money is charged to the payers when the vote is counted and credited to a recipient at the end of each round it holds the card.
Nothing else changes when a proposal passes: no stock is moved and no emissions are cut by the vote. Kept promises raise the members' trust in each other and broken ones lower it, which moves diplomatic cooperation, and an agreement the player has not broken counts with treaties and deals in the Governance score. A computer-run country votes once, when the proposal is put, the way it would answer the same terms as a two-country deal; its no cannot block the class, and if it says yes it delivers its card once, at its delivery share.
Class chat has one channel for the whole game and direct messages between two countries; only the game's players read them, and the instructor can read and remove messages. Chat, proposals and votes ride on the status request each browser already makes every 5 seconds (only new messages are sent), so they reach other players within about 5 s. A solo game has deals with computer-run delegations and treaties, and no chat or class votes.
Treaties work separately. The game holds 18 real treaties; a country begins as a member of those it has joined in reality (shown as "member since", with nothing to do in play), and 13 of them can be joined in play. A treaty is a commitment: joining cuts no emissions by itself, costs no capacity points and gives nothing at the click: every benefit comes after the obligation is met. A finance pledge's obligation is to pay its dues, every round it is held, 4% to 11% of a round's budget. Each treaty names an obligation: a matching card at 50% or more, a cut against the country's own baseline that grows each round, a research programme, or the dues. It is checked each time a round ends and must be met by the end of the round after joining. A check that is met puts the member in good standing: it adds 50% of the treaty's cooperation figure (up to the figure), counts as an agreement kept in the Governance score, and from the next round returns up to 20% of what the treaty's cards cost and builds their own cut up that much faster, never beyond it. A missed check withdraws that and costs 3 cooperation each time. Leaving is always possible, for 4 to 8 cooperation; dues already spent are not returned. No treaty changes fossil CO₂, methane or land-use carbon by itself: the Global Methane Pledge and the forest treaties commit a member to the cards that do. A membership held before the game is in the baseline. In a class game each browser reports its country's treaties and their standing to the class server, so every player sees who else is in.
11 / Scoring
After each round, country performance is evaluated across four pillars. A weighted sum produces an overall score (0–100), but the sub-pillar breakdowns are visible in the Results dashboard so players can diagnose trade-offs rather than just observe their aggregate number. The pillars interact through the game's stocks. The Climate score reads the emissions outcome directly: it is not multiplied by policy effectiveness, so the same emissions cut earns the same credit at any level of trust. Public trust and governance capacity are scored in the Social and Governance pillars. Spending that drains fiscal capacity or raises employment pressure pulls the Economic score down, and accumulated climate damage weighs on both the Social and Governance scores.
Three quarters: the country's emissions cut against its own baseline, 0.75 points for each percentage point, the same in every country, so equal proportional effort earns equal credit whatever the country's size. One quarter: 2100 warming, on a scale set by what a game can reach: nothing at the no-new-policies warming of 2.85°C (5.13°F), all of it at the stretch target of 2.4°C (4.3°F) or below, in proportion between. The result is not multiplied by policy effectiveness, and loses up to 12 points when renewables above 20% can't reach the grid. The Overview mission is 2.5°C (4.5°F) with a 2.4°C (4.3°F) stretch tier, both read from a search of whole games on the 2.85°C (5.13°F) baseline. The mission is what a class of eight reaches when every country plays strong: 2.38–2.45°C (4.28–4.41°F) on the eight screens, shown as 2.5 at the highest. The stretch is the lowest warming a class can reach, 2.35–2.41°C (4.23–4.34°F). A class of typical play ends at about 2.7°C (4.9°F), and no single country reaches the mission alone: with treaties and deals China ends at 2.62°C (4.72°F), India at 2.71°C (4.88°F), the USA at 2.74°C (4.93°F) and the other five at 2.75–2.83°C (4.95–5.09°F).
The Economy score has three parts. Value for money, 50 points: the share of your emissions you have cut, divided by the share of the round's budget your cards and research take, with at least 50% of the budget counted as spent, so one cheap card cannot earn it all. Growth kept, 35 points. A programme within budget, 15 points: the share of rounds in which you had at least one card in force and the round closed within its budget. Money left unspent earns nothing. Growth kept loses 3.9 points for each point of growth lost since the game began, so a fast-growing country has no head start. Employment pressure and fiscal capacity below 40 subtract from the score. With no policy the score is growth kept alone.
Public trust (40%), preparedness (35%) and a fixed 25-point base, minus penalties for the share of event losses that preparation left unavoided (25 points unprepared, none at the largest cut the game allows, whatever the size of the economy), the accumulated climate-damage stock (up to 20) and inequality risk above 40 (up to 10). Crisis response choices made between rounds shift trust and preparedness, making the Social score the most sensitive to inter-round gameplay decisions.
Diplomatic cooperation (up to 40 points), low diplomatic tension (up to 35; nothing in the game moves tension yet, so this part stays at 28), agreements made and kept (3.5 for each treaty joined in play whose obligation is met at the last check and each deal a partner accepted, up to 10), a fixed 5, and institutional capacity (±10 around its neutral point of 60), minus up to 10 for accumulated climate damage. The pillar reflects the empirical reality — documented across multiple IPCC AR6 chapters on international cooperation — that durable climate action at scale requires durable multilateral institutions. Countries that defect from cooperation systematically underperform on this pillar even if their domestic emission cuts are strong.
Weights (Climate 50% / Economic 20% / Social 15% / Governance 15%) are defined in src/simulation/scoring.ts. Sub-pillar breakdowns are visible in the Results dashboard after each round. The letter grades in the instructor panel are set from a search of whole games, not from a school scale: an F is below 21 (no better than doing nothing), a C starts at 25 (typical play), an A at 75 (a class in which all eight countries play strong), and the best game the rules allow scores about 81. A strong game played alone ends on 66 to 74, a B+: a quarter of the Climate score is the world's warming, which one country moves little. No letter is shown before the first round has ended.
12 / Sources
Observed baselines and climate-model parameters come from published data products and assessed science. Policy-card effects and prices are calibrated to the system-dynamics model. Governance rules and many of that model's parameters are educational estimates informed by that literature. The table below maps each data domain to its source. Publication years and observation years may differ; these are the dataset vintages used in the simulation, not a claim that they are the latest releases. For traceability of individual numbers, see the inline citations in src/data/citations.ts.
13 / Technology
The simulation runs in the browser and calls a Python model service for the climate projection, the system-dynamics step and classroom sessions. If that service is unavailable, a simplified local model keeps the game running.
14 / Design choices
The following design choices were made deliberately to balance quantitative rigour with educational usability. Each represents a trade-off: removing it would improve model fidelity but at significant cost to interpretability or runtime.
15 / Boundaries
Graduate-seminar users should treat this list as the primary caveat when interpreting any in-app quantitative output.
No full cryosphere model
AMOC and WAIS tipping points are modelled as logistic activation functions with quantitative damage multipliers, not a coupled ice-sheet or ocean circulation model. Committed sea-level rise and regional climate impacts are captured as aggregate damage signals, not spatially resolved projections.
Thin economic feedback
Policy ambition modifies sector emissions multiplicatively; it does not feed back into energy prices or output the way a full integrated assessment model (IAM) would. In the system-dynamics model GDP grows 2.76% a year at 2020 income and more slowly as income rises, less the output that unrepaired climate damage and unserved energy cost and a small drag from the cost of the policy portfolio. Policy reaches it through avoided damage and through that cost, and energy demand follows that GDP. A country’s own growth rate, which the Economic score reads, moves with crises and a few policy side effects, not with detailed macroeconomic channels.
No trade, migration, or supply-chain channels
Carbon leakage, resource competition, and climate-driven displacement are represented only narratively via crisis events, not quantitatively. Multi-country emission interdependencies are modelled only through the sum of the individual country emission pathways in the global FaIR input.
Simplified non-CO₂ and land-use forcing
2.2 includes every forcing agent. Player choices change fossil CO₂, scale scenario CH₄ and N₂O, and reduce positive land-use CO₂ through a conservative national net-flux proxy. Forest protection cannot strengthen existing net sinks in this approximation; the Forest Restoration card adds a capped removal on top, with no re-release of stored carbon. Aerosols, black carbon, ozone precursors and HFCs retain the medium-extension scenario; HFCs are not taken from players.
Mostly deterministic runs, partial uncertainty
Apart from the hidden tipping thresholds each game draws at random, a given policy sequence produces the same outputs every time it is run; crisis events come from seeded draws. The temperature projection carries climate-response uncertainty (5–95% range across a fixed 50-member subset of the 841-member constrained FaIR ensemble, picked at evenly spaced quantiles of its 2100 warming; the subset’s median is within 0.01°C (0.02°F) of the full ensemble’s), but the emissions pathway, the ODE and the policy effects are single deterministic estimates. The effect sizes on policy cards are single figures calibrated to the system-dynamics model; the ranges shown for the two adaptation cards are literature spreads (IPCC AR6 indicative ranges), not posterior uncertainty. Real projections also carry scenario and socioeconomic uncertainty that the game does not represent.
Country and world resolution
Sub-national detail, such as sector composition, grid topology and regional impacts, is collapsed into country aggregates. The choropleth and raster maps in the dashboard visualise real satellite data, but the emissions model and the scores work country by country, and the system-dynamics model is a stylised world economy driven by the selected player’s policy levers in the server step.
Sector splits
Each country’s sector shares of fossil CO₂ come from Climate Watch (2023) and are applied to its 2024 total, so they lag by a year. Climate Watch doesn’t separate farm fuel from other unspecified fuel use, so the agriculture share is an upper bound: close to FAOSTAT’s on-farm energy CO₂ for most countries, but mostly unspecified fuel for India.
Thin negotiation model
Bilateral deals are answered by a deterministic diplomacy engine: each delegation scores an offer against its sourced profile, its trust in the player and its red lines, so the same offer always gets the same reply. Multilateral proposals are thinner: in solo play the other delegations vote and propose by seeded random draws, not by their interests, and a passed proposal mostly moves cooperation.
17 / Citation
If you use Climate Futures in teaching or research, please cite both the software and the underlying data products, namely IPCC AR6, FaIR 2.2 (Leach et al. 2021) with fair-calibrate v1.4.1 (Smith et al. 2024), the Met Office HadCRUT5 and CRUTEM5 records, NOAA GML, Indicators of Global Climate Change 2025 (Forster et al. 2026) and Armstrong McKay et al. (2022), as referenced in the relevant panels above. In-app hover text on policy cards includes chapter-level AR6 pointers for individual instrument traceability.
