The Net Zero Game

How complete is the calculation engine?

Transport levers

Modal destination shares redistribute the 2020 service demand of a source mode among 2050 modes. Existing activity in other modes remains in the calculation.

Passenger or freight demand reduction is applied to all passenger-kilometres or tonne-kilometres before modal allocation.

Biofuel share splits liquid fuel between biofuel and e-fuel. E-fuel production uses electricity with a 40% conversion efficiency.

Building-heating levers

The stock — 3 655 Mm² across 24 segments, 8 heating systems × 3 building types — says how much heat the country needs and anchors the winter peak. What covers that heat is set by target, and gas has no slider: it absorbs whatever the targets leave uncovered. That is what makes the account close by construction, and what makes the cost of not choosing visible.

Biomass is a target in TWh of wood burned, not a share, so it can be read straight against the biomass limit on the dashboard instead of being reconstructed from two shares. Electrification is a share of the heat need — of heat, not of energy; how that heat is produced is the next question down, and it is where the peak is won or lost.

The five electric technologies are not interchangeable. Over a year an air-water pump returns 3 kWh of heat per kWh of electricity, an air-air or network pump 2.5, a resistance 1. On the coldest evening both air pumps fall to 2, a network pump to 1.5, a resistance stays at 1, and a hybrid moves 70% of its load onto gas while running 95% electric over the year. Electric resistance is a slider rather than a stock that can only shrink, because a scenario may genuinely install more of it: it is cheap to fit and the worst thing that can happen to the peak.

Heat networks are declared in TWh of wood and of recovered heat. With the network heat pumps set above, those say how much of a network is decarbonised; gas absorbs the rest along with everything else. Recovered heat has no emission factor and adds nothing to the peak, which makes it the cheapest thing a network can run on — and the model does not check it against the waste-heat gisement the industry module computes, so raising it far is optimistic in a way nothing here will stop you being.

Retrofit improvement is an average demand reduction across the whole stock, not the percentage of buildings renovated. It acts on the heat need before any system sees it, so it benefits every vector alike and is the only lever that lowers the peak without changing a single technology.

If the targets over-subscribe — more heat allocated than the stock needs — gas floors at zero and the surplus is reported beside the sliders rather than absorbed. A scenario that has quietly allocated more heat than exists is one whose numbers should not be trusted.

What this replaced, twice. Until v0.5.0 building heating was an aggregate fitted at a single point: three linear regressions, one COP of 3, one peak COP of 2. Its three carriers each implied a different total heat demand — 332 TWh via electricity, 366 via wood, 229 via gas — so the shares were not a partition and substitution did not conserve heat: on a path to 95% electric, 41 TWh appeared from nowhere. The 60% cap on electric heating existed to hide that. v0.5.0 replaced it with a transition of surfaces, which conserved heat properly but could only ever shrink electric resistance and split the biomass a scenario used between "leaving" and "arriving" shares nobody could add up. v0.7.0 keeps the stock for the heat need and the peak anchor, and sets the allocation by target.

What it still does not carry: domestic hot water, cooking, cooling and the specific electrical uses — this is space heating only, roughly half of what a building consumes. The allocation is national, so it cannot say that a heat network needs density and a detached house will not get one; the residential/tertiary split of each vector follows the heat need rather than a separate stock.

The winter electricity peak constraint

The peak indicator is the additional winter power demand created by electric space heating. It is the constraint that makes electrification a trade-off rather than a free win: a scenario can be excellent on emissions and still be unbuildable because it asks the power system for too much capacity on the coldest evenings.

It is built from the stock: every segment's heat need, at its system's peak efficiency rather than its seasonal one, counting only the share of that system actually running on electricity on the coldest evening. Those three things differ by technology in ways a single COP cannot express. Air-air and air-water heat pumps fall from 2.5 and 3.0 seasonal to 2.0 apiece at peak; district-heating electricity falls from 2.5 to 1.5; electric resistance is 1 in both, which is why retiring it is the strongest single lever here. A hybrid heat pump runs 95% on electricity over the year but 70% on gas at peak, so it is by some distance the cheapest way to electrify heat without buying winter capacity — and the gas shows up in the emissions.

The 2020 figure of 40 GW is an anchor, not an output: the same expression is evaluated for the 2020 and the 2050 stock and their ratio scales it, so freezing the stock returns the anchor. Thresholds are 35 GW (target) and 45 GW (limit).

The reference scenario is over the limit, at 51 GW, and that is the finding rather than a slip. The workbook's own peak formula divided by the peak efficiency twice, and anchored its 40 GW against the 2020 useful heat instead of the 2020 peak load — two different quantities, so running its own 2020 stock through it returned 36.8 GW rather than 40. Its answer, 39.45 GW, is the number the old aggregate module was fitted to reproduce, and the 35/45 band was set against it. Corrected, the transition the workbook describes does not hold the winter peak flat: it multiplies it by about 1.27. The band was deliberately left where it was, because a scenario that meets its carbon targets and still cannot be built is the thing this indicator exists to show. The levers out of the red are real ones — retrofit, hybrid heat pumps, district heating, and not replacing electric convectors with more electric convectors.

Scope limitation: only building heating is counted, as in the workbook. Electricity used by transport, industry, hydrogen and e-fuels changes the annual energy but is not added to this peak, even though electrolysers and industrial loads do interact with system adequacy in reality.

Building usages other than heating

Space heating is about half of what a building consumes. This is the other half — hot water, cooking, air conditioning, and the specific electrical uses: lighting, appliances, screens and the servers behind them. Until v0.8.0 the model left all of it out, which meant it was scoring roughly half the sector.

The data is CEREN's, as the SDES publishes it. Residential is 2024, the latest available. Tertiary is 2019, not 2020: 2020 is a Covid year in which tertiary consumption fell from 237 to 209 TWh and recovered afterwards, so using it would have built a lockdown into the 2050 baseline.

Heat-pump ambient heat is excluded. The source reports it beside the electricity that drives the pump; counting both would double the energy.

It is a weaker model than the heating one, deliberately. There is no stock and no technology choice: each usage is its observed energy carried to 2050 and moved by efficiency, growth, or both. The alternative was to leave 240 TWh out of the account entirely.

Three gaps, named. Fuel switching in hot water and cooking is not a lever — their carrier mix is carried forward as observed, so a scenario cannot electrify a gas water heater here. Air conditioning makes a summer peak and the only peak this model constrains is a winter one, so its growth costs energy and emissions but is never scored against a capacity limit. And district heat is folded into gas, the model having no heat carrier outside the heating module — 2.8 TWh, stated rather than buried.

The accounting scope — read this before comparing anything

This model is a scope-1 account. Emissions are booked where the combustion happens. A power station's emissions belong to the power station; they are not spread back over everyone who used a kilowatt-hour. Electricity therefore carries nothing where it is consumed — a building that electrifies its heating shows zero emissions for that electricity, and the emissions appear in Electricity generation instead, computed from the fuel the chosen mix actually burns.

The consequence to hold onto: electrification moves emissions rather than removing them. Where they land depends on the electricity mix, which is a separate choice on the Supply tab. Grow the specific electrical uses by half and building emissions do not move at all — the energy sector does.

This is the convention SECTEN and the SNBC use, which is why the national reconciliation is now a short page: the only difference left between the two accounts is international aviation and shipping, which the inventory reports as a memo item outside the national total.

Until v0.11.0 it was a footprint account — every sector charged the life-cycle emissions of its electricity at 40 gCO₂/kWh. That is a legitimate convention and it answers a different question: what does this sector cause? rather than what does this sector burn? It made the game total 61.3 MtCO₂ where it is now 30.0, and it made buildings 15.6 where they are now 5.0. Neither number is wrong; they are answers to different questions, and the model now answers the one the national inventory asks.

What the grid factor is and is not. The model derives it from the mix rather than declaring it: roughly 1.7 gCO₂/kWh at the reference. That is a combustion figure — no construction, no fuel chain, no decommissioning — so it is not comparable with the 80-odd gCO₂/kWh a life-cycle study reports for the same grid. Comparing the two is the most common way to make this model say something it does not say.

The electricity mix

The supply follows the demand rather than standing beside it: whatever electricity the rest of the model turns out to need is served by the share structure of one of RTE's six 2050 scenarios, from M0 at 100% renewable to N03 at about half nuclear. Choosing a scenario answers with what, never how much.

Capacity follows from energy through a load factor — RTE's own, read back out of its capacity and generation tables, and remarkably stable between scenarios: onshore wind 23%, offshore 41%, solar 14%. What has to be built each year follows from capacity through a lifetime, on the reasoning that a fleet of that size has to be renewed at that rate. It is a build rate, not a build programme: it understates the years the fleet is still growing and overstates them once it is not.

Two splits the source does not make are made here. Solar is halved between ground and rooftop, offshore wind between fixed and floating. Both matter to the material account — a floating foundation is 480 t of steel per MW against 250 fixed — and neither is a result.

This does not check that the mix works. There is no hourly balance, no adequacy calculation and no storage: a 100%-renewable share structure is applied exactly as a nuclear-heavy one is. The winter peak the building module computes is a demand-side number that nothing on this side has to meet, and the whole question of what a renewable-heavy mix costs in flexibility is absent. The cost shown is plant only — capital recovered over each technology's own life, plus fixed operating cost. No fuel, no carbon, no network, no storage. It lands near RTE's own figure for its scenarios, which is reassuring about the arithmetic and says nothing about the omissions.

How the hydrogen is made

One mix serves every hydrogen consumer in the model. Three routes: electrolysis, which buys hydrogen with electricity at the 60% conversion the rest of the model uses; steam methane reforming, which buys it with methane; and autothermal reforming with capture, which does the same and puts 94% of the carbon underground — ATR concentrates the CO₂ in one stream, which is why it captures where a reformer with post-combustion capture struggles past 60%.

Ammonia no longer owns a route. It used to be two rows — 700 kt from electrolytic hydrogen, 200 kt from a reformer — which put the hydrogen decision inside the ammonia lever and nowhere else. Now every tonne consumes the same 5.94 MWh of hydrogen and the mix decides how it was made, which is where that decision belongs: the same reformers serve steel and everything else.

The capture credit is charged against the physical carbon, not against the emission factor. Those are different numbers and both are needed: efGas at 25 gCO₂/kWh answers "what does burning this count as?", while carbon_in_methane at 202 gCO₂/kWh answers "how much carbon is there to capture?". A capture plant removes molecules, not conventions.

Hence the negative number, and hence the warning. Reforming biomethane with capture takes carbon out of the air and puts it underground, so the route reads about −18 MtCO₂ a year at full deployment — enough to close the gap to the SNBC on its own. That is the physics of BECCS. It is also the point at which this model will most easily mislead: it says nothing about whether the biomethane exists, what land it came from, or whether the storage holds. The Controversy tab says so too.

What is missing. No separate capital cost for the capture train — the ATR route uses the reformer's annuity, which understates it. No transport or storage cost for the CO₂. And the methane a reformer needs is charged to the biogas pool, which at full reforming is well past anything France could supply.

Materials of the transition

A decarbonisation pathway is usually argued in TWh and MtCO₂. This says what the same pathway weighs: the steel, concrete and critical metals it asks for each year in 2050.

It is a satellite account, and deliberately a one-way one. It reads the scenario; nothing reads it back. The steel a wind farm needs is not charged to the steel industry the model already has, the concrete is not charged to cement, and none of it emits. Wiring it back would double-count against an industry whose output is set by its own levers — so the honest thing is to compute the demand and put it beside the supply rather than inside it. A test pins that no material lever moves emissions, energy or cost.

What drives it. Generation is a declared build rate in MW per year, because the model has no electricity supply module: it computes demand, not a mix. Vehicles are a declared annual production, but the share of it carrying a battery follows the player's own electrification levers for cars and trucks. Those levers are shares of demand rather than of production; over thirty years the two converge, and the approximation is stated rather than hidden.

The chemistry lever is the sharpest trade-off here. LFP carries almost no cobalt — 7 grams per MWh against 27 kilogrammes — and a quarter of the nickel, but 4.4 times the lithium, 490 kg per MWh against 111. There is no chemistry that is cheap in every metal at once.

Two comparisons worth reading. The transition's steel against the steel this scenario's own industry produces: both sides move with the player, so electrifying harder raises the steel needed and, if the industry levers are left alone, does not raise the steel made. And its concrete against clinker — a ratio above one would not be an error, since concrete is mostly aggregate.

What is missing, and it is named rather than filled. Heat pumps are absent: no source in hand gives their material content per unit, and inventing one would put a number in the annex that nothing supports. Closing it needs a per-unit steel, copper and refrigerant intensity from an LCA or from ADEME. Flat glass, plastics and rubber are carried by the source for vehicles but not totalled here. Nothing is recycled: this is primary demand, so a scenario with a serious secondary-metal loop would need less than the account says.

Industry levers

Steel production change is the relative change in 2050 steel output compared with the workbook’s 2020 route volumes. The coefficient is used as 1 + g: +30% means a multiplier of 1.30, while −20% means 0.80.

H-DRI steel share splits primary steel between the BF-BOF and hydrogen direct-reduction routes. Recycled EAF steel is scaled by the same production-change coefficient.

Green ammonia is entered in kt/y. The workbook reference also contains 200 kt/y of grey ammonia; its treatment is documented as an accounting limitation.

CO₂ + H₂ olefins combines a new production-route share with plastic-demand reduction and an optional biogenic-CO₂ credit.

Clinker ratio and capture separately affect cement production-process emissions.

National reconciliation — method

The game and the national inventory do not measure the same thing. Reconciling them by a ratio, as earlier versions did, transfers relative change but hides two differences and any sub-sector the game does not model. Each difference is now its own line.

Both accounts are now scope 1. Since v0.11.0 the game books emissions where the combustion happens, which is what SECTEN and the SNBC do: power-station emissions sit in the energy branch, at stack level, and not in the sector that used the kilowatt-hour. The line that used to undo a life-cycle electricity factor is therefore zero, and kept only so the change is visible rather than silent. See the scope section above for what that convention costs and what it buys.

International bunkers. International aviation and maritime shipping are in the game and are a memo item outside the national inventory total. The deduction is computed from the model's own international rows, so it follows the scenario instead of being asserted.

The coverage gap. The game models five industrial value chains. Everything else the inventory calls industry is a named line whose size is the difference between the official industry total and what the five chains represent — see the constant industry_covered_2020 in the generated annex for that derivation and its two caveats.

First-order sectors. Agriculture, waste and energy production interpolate linearly between observed 2024 and the current SNBC 3 2050 order of magnitude. At 100% they sit exactly on the SNBC value, so those three rows are an input, not a result.

Carbon sinks. Natural and technological sinks are separate. The −43 MtCO₂e technological value closes the ministry's −66 MtCO₂e total 2050 absorption figure after subtracting the stated −23 MtCO₂e natural sink.

Emission factors — today and in 2050

These were editable assumptions in the source workbook and are restored here as levers, because they are scenario choices rather than measurements, and because in a decarbonised pathway they decide almost everything that is left.

Aviation — method, sources and what it leaves out

The ticket is built from the bottom: the energy the model already charges the flight, multiplied by a synthetic-fuel price, plus everything else derived from today's economics.

Distance comes from the DGAC's 2023 traffic statistics for flights departing France: passenger-kilometres divided by passengers, category by category. It is an average over each category, so "Paris ↔ international" blends a Barcelona hop with a Tokyo sector.

Energy is the model's own aviation consumption, not a separate figure — 0.24 kWh per passenger-kilometre for domestic flights and 0.19 for long-haul. Be careful with these: the raw French statistic for 2024 is 29.3 g of kerosene per passenger-kilometre, about 0.35 kWh, because it also carries the freight in the holds and reflects actual load factors. Corrected for both, the same series gives about 19.3 g/pkm, which is the range the model sits in. A real ticket therefore emits more than the figure in the table.

Fuel price. The published estimates for sustainable aviation fuel disagree by a factor of six, and the review behind the two sliders spans EASA, the European Commission's ReFuelEU impact assessment, the DGAC roadmap, ISAE-Supaero, ATAG's Waypoint 2050, E-Cube, the IEA, Solakivi et al. (2022), Brynolf et al. (2020) and Massol et al. (2025). Bio-jet from waste oils is the cheapest route at 600–1 900 €/t; power-to-liquid the dearest at 1 800–10 000. The defaults sit mid-range and the ranges are the honest answer, which is why they are sliders.

Efficiency. The source workbook gives 2050 aviation the same consumption per passenger-kilometre as today, so the default gain is zero. The published trajectories converge on about 1%/year — ICAO, ADEME, T&E, the UK Committee on Climate Change and the World Economic Forum's Clean Skies for Tomorrow all sit between 0.9 and 1.0. Moving the slider to 1 compounds to a 23% saving over twenty-six years, which is less than most people expect and is the point of exposing it.

What this omits

Non-CO₂ effects — contrails and nitrogen oxides — which several studies put at the same order of magnitude again as the combustion CO₂; the upstream chain of the fuel; any change in airline cost structure between now and 2050; airport and air-traffic-control investment; and the question of whether the biomass or the electricity these fuels need is available at all. The last one is not rhetorical: the scenario's aviation fuel alone is a large share of the biofuel pool the scoreboard already flags.

The traffic categories, the fuel-cost review and the efficiency trajectories were assembled in a working file that is not public. Only the derived parameters appear here, each attributed to its primary published source.

The rest of industry — how output and processes were separated

The seventeen branches below were, until this version, a single residual line in the national reconciliation: a number added back to close the account, with no energy behind it and no lever on it. They are now modelled.

Why two levers and not one

The source scenario for 2050 is not a projection. It electrifies — and it also grows output, by very different amounts branch by branch. Measured from the branch sheets' own production data: textile ×8.5, electronics ×3.1, mineral extraction ×2.5, diverse industries ×1.8, rubber ×1.8, pharmacy ×1.5, vehicles ×1.5, machinery ×1.3, non-ferrous metals ×1.1; foundry, ceramics and glass unchanged; paper ×0.86, plastics ×0.70, plastic products ×0.67, naval and aerospace ×0.56, mineral chemistry ×0.52. A single slider would let a player appear to clean up industry while quietly assuming an eightfold textile sector, so output and process are separate.

The decomposition is exact, not fitted

Energy is output × unit consumption, so it is bilinear in the two and four corners reproduce every combination exactly: E00 the observed 2019 situation, E11 the source scenario, E10 the 2050 output at 2019 processes, E01 the 2050 processes at 2019 output. Both end points are the published branch totals. The output index between them is measured, branch by branch, as the energy-weighted ratio of 2050 to 2019 production over the product rows that reproduce their own computed total — between 75% and 100% of each branch's energy, and 100% for thirteen of the seventeen.

At today's output, the 2050 processes take the electricity of these branches from 68 to 138 TWh while cutting coal and fuel oil to zero. Output growth alone would take it from 68 to 87. Both together give the source scenario's 166.

Conventions and what is still missing

Purchased steam is carried with gas, non-renewable waste fuel with coal, and residual fuel oil with the model's liquid-fuel carrier, which in 2050 is biofuel or e-fuel by the same assumption the transport module makes. Process emissions — glass, other building materials, mineral chemistry, 2.5 MtCO₂ in 2019 falling to 1.9 — follow the same two levers.

Not represented: any lever finer than the branch group. Cross-cutting energy efficiency and waste-heat recovery both have published ceilings that would fit here — RTE puts the electricity-efficiency potential at 9.0% for paper to 31.1% for chemistry at long payback, and ADEME finds 15.6 TWh of recoverable waste heat, 7.7 of it above 100 °C — but neither is in the model yet, because applying them correctly needs the temperature split that exists at process level in the source and has not been aggregated. Stated rather than approximated.

Waste heat — a resource that decarbonisation consumes

Industrial processes reject heat. Some of it can be recovered and used instead of burning more fuel. The usual way to model this is a fixed reserve in TWh, and that is wrong in a way that matters here.

Why the gisement shrinks

Waste heat is a by-product of combustion and of process inefficiency. An electric furnace or a heat pump rejects far less of it, and at lower temperature. So the more a scenario electrifies industrial heat, the smaller the waste-heat resource it has left to recover. Recovering waste heat and electrifying heat compete for the same physics, and a model that treats the gisement as a constant lets a player count the same energy twice.

ADEME's study makes the coupling possible because it expresses the gisement against the fuel each sector burns, not as a bare total: 15.6 TWh recoverable on the 2019 industry, 7.7 of it above 100 °C, out of 248 TWh of fuel — 6.3% on average, but from 1.2% in metals to 31% in paper, where drying dominates. The model attaches those intensities to the fuel, post by post, so the gisement follows whatever the scenario actually burns.

Conventions

Recovered heat displaces gas, the marginal fuel, and cannot displace more than the post burns. The second-order feedback — less gas means a slightly smaller gisement in turn — is neglected; at full recovery it is under half a percent. Transport and buildings carry no gisement because the ADEME study is industrial. Glass is listed by ADEME under both chemistry and non-metallic materials; it is assigned to materials here, its furnaces being the hotter of the two contexts.

What this does not say

That the heat can be used where it is produced. Above 100 °C it can displace process heat directly; below, it needs a heat pump to upgrade it or a district network to carry it somewhere useful, and neither is costed here. Roughly half the gisement is below 100 °C, so a recovery rate above 50% implicitly assumes one of those. Nor is the capital cost of recovery represented anywhere in the cost layer.

Energy efficiency — a ceiling, not a wish

Motors, drives, compressed air, insulation, heat integration: the cross-cutting savings that need no change of process. The danger with a lever like this is that it lets a player invent efficiency, so it is bounded by what a published study actually found.

The ceiling

RTE, after CEREN, identifies a potential of 21.1% on industrial electricity and 19.8% on fuels, of which about 58% pays back in under three years. The electricity ceiling is branch-specific and applied as such, which matters because the spread is wide:

The lever says how much of that identified potential is captured, not how much exists. At 100% every branch reaches its own ceiling and no further; at about 58% the scenario is taking roughly the part that pays back in under three years, which is the honest "no-regrets" anchor to argue from in class.

Efficiency also destroys waste heat

Fuel that is never burned rejects no heat. So the efficiency effort shrinks the waste-heat gisement exactly as electrification does — in the reference scenario, from 10.7 TWh to 8.6 at full effort. Efficiency, electrification and waste-heat recovery all draw on the same combustion, and the model makes them compete rather than letting a scenario bank all three.

What is not represented

Only direct electricity carries the electricity ceiling: the electricity that goes into hydrogen and e-fuel is governed by conversion efficiencies declared elsewhere. RTE gives no branch breakdown on the fuel side, so one ceiling applies across industry. Transport and buildings are untouched, their own levers already carrying demand and equipment efficiency. And nothing here costs the investment that buys the efficiency — the cost layer prices energy and plant, not retrofit of motors and heat exchangers.

Every input, with its provenance

This annex is generated from the model specification itself — model/parameters.yaml and model/equations.yaml — so what is documented here and what the engine executes are the same thing. The model has 69 levers, 76 constants, 20 data tables and 304 equations.

Levers — what the player can move

Constants — fixed inputs

Data tables

The categories the model iterates over. Every row is addressed by its identifier, which is what the formulas in the next section refer to.

Passenger transport categories Workbook

Demand is 2020 service demand in billion passenger-kilometres; unit consumption is per vehicle-kilometre and occupancy converts it back to passenger-kilometres. `in_inventory` says whether the category is inside the national inventory perimeter: international aviation is reported by SECTEN as a memo item and excluded from the national total.

• Teaching workbook, "Transport parc 2020" and "Transport parc 2050" sheets
• Citepa, SECTEN 2026 — international bunkers are excluded from the national total

Passenger demand reallocation Workbook

Each row moves a share of one 2020 category's demand to a 2050 category. Shares that a lever drives are overridden in the equations; the rest are fixed workbook conventions. A category with no row keeps nothing, which is how fuel cars, fuel utility vehicles, fuel two-wheelers and fuel buses are retired.

• Teaching workbook, "Transport parc 2050" sheet

Freight transport categories Workbook

Demand is 2020 service demand in billion tonne-kilometres and unit consumption already includes loading. International air freight carries the gas vector in the workbook, so about 23 TWh of the game's "biogas" resource is in fact air-freight fuel — a workbook convention worth knowing before reading the biomass scoreboard.

• Teaching workbook, "Transport parc 2050" sheet

Freight demand reallocation Workbook

Building heating systems Workbook

The eight systems the stock is made of. `is_destination` says whether new surface can arrive: fuel boilers and electric resistance can only shrink, because nobody installs either in 2050, so the six remaining systems are what the stage-2 mix distributes over. `heat_pump` is used by the cost layer to price the equipment actually installed.

• Teaching workbook, "Building parc 2020" and "Building heating" sheets

Building stock segments, 2020 Workbook

3 654.9 Mm2 and, after the stock-wide calibration, 359.34 TWh of heat need in 2020. Surfacic need is what the segment asks of its heating system per square metre and per year, before retrofit; it is a *need*, not a consumption, so the efficiencies in `building_vector` have not been applied yet. `dwellings` is informational and no equation reads it: the workbook counts tertiary floor area directly rather than in buildings, so it is zero on those three rows.

• Teaching workbook, "Building parc 2020" sheet rows 3-26

Heating system efficiencies and vector mix Workbook

How a segment's heat need becomes energy: need x unit consumption / efficiency, summed over the pairs below. A system can draw on several vectors -- district heating on six, a hybrid heat pump on two -- and `unit_2020` / `unit_2050` are the repartition keys, which is why they sum to one per system rather than carrying a physical unit. Two columns carry the whole argument for replacing the old aggregate. `peak_efficiency` is separate from `seasonal_efficiency` and lower for every heat pump (air-air 2.5 -> 2.0, air-water 3.0 -> 2.0, district-heating electricity 2.5 -> 1.5), because a heat pump loses efficiency exactly when the system needs it most. `peak_share` makes the hybrid heat pump run 70% on gas on the coldest evenings while running 95% on electricity over the year. The module this replaces flattened all of it into two constants, a seasonal COP of 3 and a peak COP of 2.

• Teaching workbook, "Building heating" sheet rows 22-35

Industrial production routes Workbook

Unit consumption per tonne of product. The cement row carried electricity and a process term but **no kiln fuel at all** until 0.9.0: a clinker kiln burns 0.70 MWh a tonne and the model had it burning nothing, which left about 9 TWh of industrial fuel — and the combustion emissions that go with it — outside the account. The three fuel intensities are the source workbook's own, from its plaster/lime/cement branch sheet.

• Teaching workbook, "Industry" sheet rows 41–54

Industrial plant, capital and fixed cost Published

CAPEX in euros per tonne of annual capacity, lifetime in years, fixed O&M in euros per tonne of capacity per year. The annualised cost is CAPEX × CRF(discount rate, lifetime) + fixed O&M, at full utilisation.

• POMMES-INDUSTRY, France 2050 — conversion_investment.csv and conversion_operation.csv

Flight categories Published

The DGAC's own route categories for traffic departing France in 2023, with passengers and passenger-kilometres, from which an average distance follows. `game_row` says which of the model's three aviation categories supplies the unit consumption, so the ticket is priced on exactly the energy the emissions account already charges.

• DGAC, Bulletin statistique du transport aérien commercial 2023

The rest of industry — energy and process emissions Published

The seventeen manufacturing branches the game does not model as value chains, grouped so the published efficiency and waste-heat studies map onto them. Energy is output times unit consumption, so it is bilinear in the two levers and four corners reproduce every combination exactly: e00 is the observed 2019 situation, e11 the source workbook's 2050 scenario, e10 the 2050 output at 2019 processes and e01 the 2050 processes at 2019 output. Both end points are the published branch totals; the output index between them is measured from the branch sheets' own production data, not assumed.

• INSEE, enquête EACEI 2019 (via Offre_et_demande.xlsx, 'Consommation énergie 2019')
• Offre_et_demande.xlsx — branch trajectories to 2050 and process emissions by branch

Building energy by usage, observed Published

Final energy by usage and by carrier, from the CEREN series the SDES publishes. Residential is 2024, the latest available; tertiary is **2019 rather than 2020**, because 2020 is a Covid year — tertiary consumption fell from 237 to 209 TWh and recovered afterwards, so using it would build a lockdown into the 2050 baseline. The heat harvested by heat pumps is excluded: the source reports it as a renewable input beside the electricity that drives the pump, and counting both would double the energy. `heat` is district heat, which the model has no carrier for and which the equations fold into gas — networks in this model are majority gas, so it is the least wrong of the available homes for 2.8 TWh, and it is stated rather than buried. Space heating is deliberately absent: it is the stock model, and the two perimeters do not match — this table's heating rows would say 335 TWh against the stock's 383 for 2020, different years and different methods.

• SDES, consommation d'énergie du secteur résidentiel 2024 (données Ceren)
• SDES, consommation d'énergie du secteur tertiaire (données Ceren), année 2019

RTE 2050 generation mixes Published

Each row is one of RTE's 2050 scenarios reduced to shares of total supply, so it can be applied to whatever electricity this model's own demand turns out to be rather than carrying RTE's demand with it. The range is the one RTE built it to span: M0 has no nuclear at all, N03 about half. Two splits the source does not make are made here and declared: solar is halved between ground and rooftop, and offshore wind between fixed and floating. Both matter to the material account — a floating foundation is 480 t of steel per MW against 250 fixed — and neither is a result.

• RTE, Futurs énergétiques 2050, chiffres clés par scénario, horizon 2050

Generation technologies, material intensity Workbook

Tonnes of material per MW of capacity built, 2050 values. These drive a satellite account: the model has no electricity supply module, so what is built here is a declared build rate rather than a mix sized to cover the demand the model computes. Making it cover that demand needs load factors the source does not provide — that is the next step, and until it is taken these numbers say what a build costs in materials, not whether it is enough. `thermal_efficiency` and `fuel_carrier` are what make the account scope 1: a thermal plant burns a fuel, that fuel goes through the same constructive account as every other, and it takes the emission factor of its carrier. A combined cycle running on biomethane therefore emits at 25 gCO2/kWh of fuel and not at 356 — and it draws on the same biomethane the buildings want, which is a competition the model did not previously represent. Efficiencies are RTE's own: 60% for a combined cycle, 25% for bioenergy. **The bioenergy plants burn gas here, not wood.** RTE's category mixes solid biomass, biogas and the renewable part of waste; sending it to wood would have the power system eating 54 TWh of a resource the buildings and industry are already short of, to make electricity at 25% efficiency, which is the one thing biomass should not be used for. Calling it biogas is a modelling choice and it is deliberately conservative on quantity: a gas engine would be nearer 40% efficient than 25%, so the fuel this asks for is on the high side rather than the low. The gas turbine runs on hydrogen in every RTE 2050 scenario, so it burns no fuel here; the electricity that made the hydrogen is not traced back, which is a hole and a small one at these volumes. Hydro, bioenergy and the two thermal rows carry intensities but no lever: the source scenario builds none of them after 2050, and a slider that only ever sat at zero would be decoration.

• Offre_et_demande.xlsx, "Energie" sheet rows 19-33 — intensité matière 2050

Vehicle production and material intensity Workbook

Kilogrammes of body material per vehicle and units produced per year in 2050. `electric_share` is the share of that production carrying a battery; it is derived from the source's own battery-capacity row rather than assumed, and it is overridden by the player's own electrification levers for cars and trucks. Only steel and aluminium are carried here. The source also gives flat glass, plastics and rubber per vehicle, which are real but are not what the transition changes — the battery is.

• Offre_et_demande.xlsx, "Matériaux transports" sheet rows 2-9 and 32-33

Battery material intensity by chemistry Workbook

Tonnes per MWh of battery. The two the model blends are the two credible 2050 options; the source also documents NMC 333, NCA and LTO.

• Offre_et_demande.xlsx, "Matériaux transports" sheet rows 19-28

Hydrogen production routes Published

Per MWh of hydrogen produced. `methane` is the feedstock and fuel together, which is how a reformer is measured; `carbon_captured` is the share of the carbon in that methane that ends up underground rather than in the air. Electrolysis has no methane, and its electricity column is zero because the figure is derived in the equations from the conversion efficiency the rest of the model uses — declaring it here as well would be two copies of one number. The reformer figures come from the constants the cost layer already used and nothing else did: 3.33 t of methane and 9.23 tCO2 per tonne of hydrogen, converted at the lower heating values declared beside them.

• IEA, The Future of Hydrogen (2019) — SMR and ATR reference plants
• Teaching workbook, "Industry" sheet — SMR methane and CO₂ per tonne of H₂

Scoreboard bands Game rule

These set the difficulty of the game; they are not resource assessments from a published national study, and the biomass one in particular is argued over — see the controversy table. Declared here rather than in the interface because the feasibility tool has to score the same scenario the player does. Two copies of a rule are two rules. The four emission bands are on the *game perimeter* — the footprint basis — not on the national figures the sector cards show. The two differ by the electricity life-cycle and by international bunkers, so scoring one against the other's band would be meaningless.

Contested assumptions Game rule

A model that shows its sources still hides which of them are argued over. This table names them. `weight` is how much the answer moves: high means a reasonable person taking the other side gets a materially different 2050. Nothing here is a secret — every one of these is visible in the annex — but a reader should not have to reverse-engineer which numbers are settled and which are live.

Emissions and energy posts Derived

`waste_heat_share` is the recoverable waste heat ADEME finds per unit of fuel burned, and `waste_heat_hot_share` the fraction of it above 100 °C. They are attached to the fuel, not to the sector, which is the point: heat that is a by-product of combustion disappears when the combustion does. Transport and buildings carry zero because the ADEME study is industrial. The constructive account. Every emission the game reports is built up from these posts, and every post is energy times an emission factor plus a named process term. Nothing is added at the sector level that is not in this table, which is what makes a missing sub-sector visible.

Every equation

The complete calculation, in the order it is evaluated. A name in a formula is either a lever, a constant, or another equation in this list; row.x is a field of the row being computed; and sum(table.column, condition) totals a column over the rows that satisfy the condition.

Transport — demand reallocation

Every 2020 service demand is reallocated to the 2050 categories through an explicit matrix. Reading the matrix is the only way to see that, for instance, car demand shifted to rail is then served at the occupancy and unit consumption of a train rather than a car.

Transport — energy by vector

Liquid fuel is split between biofuel and e-fuel by the biofuel-share lever; the e-fuel half and all hydrogen are converted back into the electricity needed to make them, at the declared conversion efficiencies. That is why an electrified transport scenario still shows a large electricity demand even where no vehicle is plugged in.

Building heating

The stock says how much heat the country needs and anchors the winter peak. What covers that heat is set by target: so many TWh of wood, such a share of the need on electricity, and that electric heat split across five technologies with genuinely different efficiencies — in season and, which is what the peak cares about, on the coldest evening. Gas is the residual. It is not a target and has no slider: it absorbs whatever the other choices leave uncovered, which is what makes the account close by construction and what makes the cost of *not* choosing visible. If the targets over-subscribe the need, gas goes to zero and a surplus is reported rather than silently absorbed.

Industry — production volumes

Industry — energy and process emissions

Energy is production times unit consumption. Process emissions are the part that no change of fuel can remove: the carbon of the limestone, the carbon locked into the product, and the residue of the blast-furnace route once its coal has been counted as energy.

The rest of industry

Seventeen manufacturing branches the game does not model as value chains — metals and machinery, minerals, the rest of chemistry, paper, and a diverse remainder. Together they are about 174 TWh today, 68 of it electricity, more than the five modelled chains use between them. Output and processes move on separate levers because the source scenario mixes the two: it electrifies, and it also multiplies textile output by 8.5.

The constructive account — energy and emissions by post

One row per sub-sector, one column per energy carrier. Sector totals are sums of this table and nothing else. Electricity is kept in three columns — used directly, used to make hydrogen, used to make e-fuel — because the three have very different implications for the power system even though they carry the same emission factor.

Sector and resource totals

National reconciliation

Since v0.11.0 the game and the inventory share an accounting scope, and this module has much less to do. Both are **scope 1**: emissions are booked where the combustion happens, so a power station's emissions belong to the power station and not to everyone who used a kilowatt-hour. One difference remains, and it is real rather than conventional: the game includes international aviation and shipping, which the inventory reports as a memo item outside the national total. That is subtracted as its own named line. What is left is the perimeter the model does not cover at all — refining, fugitive emissions, and the sub-sectors nobody has modelled — and it stays visible rather than being divided away.

Annualised cost layer

Real euros, no inflation, no subsidy or transfer, at full utilisation of installed capacity. For every asset the annualised cost is CAPEX × CRF(rate, lifetime) + fixed O&M + Σ(input intensity × price) + on-site CO₂ × carbon price. The governing principle is that the cost layer prices the quantities the game already shows: it never substitutes a different intensity, so where the physical description of a chain is incomplete its cost is understated by the same amount.

Aviation — the price of a ticket

What decarbonised flying costs the passenger. The fuel side is computed from the same energy the emissions account charges, at a synthetic-fuel price the player sets; everything else — aircraft, crew, airport charges, maintenance — is derived from today's ticket through the fuel share of airline operating cost and held constant. That last assumption is the weak one, and it is stated rather than buried: a 2050 airline may have a different cost structure and nothing here models it.

Building usages other than heating

Space heating is about half of what a building consumes. This is the other half: hot water, cooking, air conditioning, and the specific electrical uses — lighting, appliances, screens, and the servers behind them. It carries no stock and no technology choice; each usage is its observed energy carried to 2050 and moved by an efficiency lever, a growth lever, or both. That is a weaker model than the heating one and deliberately so: the alternative was to leave 240 TWh of building energy out of the account entirely, which is what the model did until 0.8.0.

Electricity supply

The mix follows the demand rather than standing beside it: whatever electricity the rest of the model turns out to need is served by the share structure of one of RTE's six 2050 scenarios. Choosing a scenario answers "with what", never "how much". Capacity follows from energy through a load factor, and what has to be built each year follows from capacity through a lifetime — a fleet of that size has to be renewed at that rate, and it is the build rate rather than the standing fleet that consumes materials. The result feeds the material account, which is why the seven build-rate sliders it used to carry are gone. **This does not check that the mix works.** There is no hourly balance, no adequacy calculation and no storage: a 100%-renewable share structure is applied here exactly as a nuclear-heavy one is. The winter peak the building module computes is still a demand-side number that nothing on this side has to meet.

Materials of the transition

A satellite account, and deliberately a one-way one: it reads the scenario, nothing reads it back. The steel a wind farm needs is not charged to the steel industry the model already has, the concrete is not charged to cement, and none of it emits. Wiring it back would double-count against an industry module whose output is set by its own levers, so the honest thing is to compute the demand and put it beside the supply rather than inside it. What it is for: a decarbonisation pathway is usually argued in TWh and MtCO2. This says what the same pathway weighs. Three of the numbers are worth reading against the industry module directly — the transition's steel against French steel output, its concrete against French cement.

Cost layer — method and sources

The cost layer prices the physical flows the game already computes. It never uses a different quantity from the one shown in the emissions dashboard: if the physical description of a chain is incomplete, its cost is understated by the same amount, and that is stated rather than patched.

Convention

Real euros, no inflation, no subsidy or tax transfer. Annualised cost = CAPEX × CRF(rate, lifetime) + fixed O&M + Σ (input × price) + CO₂ × carbon price, with CRF(r, n) = r / (1 − (1+r)−n) and full utilisation of installed capacity. Two discount rates are exposed because an industrial investor and a household do not face the same cost of capital: moving the residential rate from 4% to 8% raises the building indicator by roughly a third, entirely through the retrofit annuity.

Where the cost numbers come from