How it works

Which plants run, and why

The simulation advances in steps of fifteen minutes whatever speed the clock is showing, and each step rebuilds every machine’s offer from scratch. Must-run output is taken first. Wind and solar come next, in full, at whatever the weather is giving. The load still uncovered is met from a stack sorted cheapest first, and storage bids into that stack on its own opportunity cost rather than waiting at the back of it. A unit pinned to Reserve is sorted last whatever it asks, with no invented price written into its bid.

What a plant bids is the cost of its next megawatt-hour and nothing else: the fuel price divided by the plant’s efficiency, plus its carbon intensity times the carbon price, also divided by efficiency, plus variable operating cost. The efficiencies are the catalogue’s, raised by research and capped at 95 per cent. A combined cycle is catalogued at 59 per cent, a supercritical coal set 45 per cent, an open cycle gas turbine 36 per cent, a reactor 34 per cent. The baseline world opens with gas at $22 per MWh thermal, coal at $9, uranium at $5 and municipal refuse at minus $4, which is a gate fee rather than a price. Gas carries 0.202 tonnes of carbon dioxide per MWh thermal and coal 0.34. The internal carbon price on the policy desk runs to $300 a tonne, moves the running order only and appears on no invoice.

Minimum load is why the running order is not simply a sorted list. A combined cycle cannot sit below 30 per cent of its rating, a reactor below 55 per cent, a refuse plant below 60 per cent, and that rating is what the machine can do in its present condition rather than its plate. So the stack is walked twice. The first pass considers only machines whose floor fits inside the load left to serve; a cold machine that would overshoot is held back, then started afterwards, smallest floor first, because by then the alternative is customers in the dark. A refuse plant is must-run because the rubbish arrives regardless, and any unit can be pinned must-run by hand.

In surplus, storage is filled first, then renewables are curtailed, then thermal plant is backed down dearest first as far as its must-run floor. Whatever the fleet still cannot stop making is spilled into dump banks and billed at $30 per MWh, and the price is capped at $2 for that step because nothing was sorted and nothing was marginal. Short the other way, the battery’s emergency reserve is released, a hydrogen turbine buys fuel if the policy allows it, and what is left over is unserved energy. Nothing comes in from outside: there is no interconnector, and nobody to sell a surplus to.

Ramp rates are not modelled. Every plant in the catalogue carries a ramp figure and nothing in the simulation reads it, so a machine already on the bars reaches whatever it is asked for inside the interval.
Unit commitment is not modelled either. Start times, start costs and minimum down times are catalogued for each plant and left deliberately unread, so the mechanic can return without the fleet being measured again.
Because a restart is free, a plant backed down in a surplus comes straight back when it is wanted, and the only thing holding a machine on the bars through a surplus is a must-run instruction.
The clearing price is a readout, and its 48 hour trailing average is what storage prices its own energy against. Each machine’s lifetime revenue is attributed at the retail tariff instead.

Demand and money

Load is built from the ground up. Every developed tile behind a 20/0.4 kV station carries dwelling-equivalents, each standing for 3.23 real dwellings, and each unit draws 1.15 kW of average electrical demand where the ground is residential, 2.35 kW commercial and 4.4 kW industrial, before space heat or car charging. Nine end-use shapes are mixed across twelve territory characters, so a commuter suburb, a business quarter and a two-shift works are different curves because they are made of different things. The shape for a street of houses that empties on weekdays peaks at hour 19, at 1.95 times its own daily mean; a process that never stops peaks at 1.07.

Heating starts below 15.5 °C outside air and cooling above 22 °C, so the seasonal shape follows the colony’s own weather. The standard world compounds demand growth at 2.2% a year, and every town’s population is restated once a month from how well it was served, public opinion and energy poverty. The opening town is founded with 9,000 to 11,500 people. Another settlement arrives three to five years in, and every four to eight years after that.

The player files one flat tariff, opening at $118/MWh, and can also offer an hourly contract at the wholesale price plus a margin. Customers judge the price against the cost of supply plus 22%, reading a slow average of what has been charged rather than the figure set this instant, and cut consumption on an elasticity of that ratio raised to the power minus 0.34. Above about 0.95 of what local incomes carry they fall into hardship and stop paying: at full hardship 42% of what is billed is never collected. A tariff held far above the cost of supply brings a fine and a five-year cap.

Revenue is delivered energy times that price times the collection rate. Against it go fuel, carbon, variable and fixed O&M, grid upkeep, disposal of unsold generation at $30/MWh, and interest. Credit is one fixed ceiling, $120 million at 4.5% on the standard world against $180 million of opening cash, the same figure on the first morning and the last, with no credit rating and no automatic repayment. An empty account under a fully drawn line ends the run in administration.

Fuel at the power station gate, dollars per MWh thermal at the founding: coal $9, gas $22, oil $45, uranium $5, biomass $15, hydrogen $120, and waste at minus $4, which is a gate fee paid to take it away.
Each fuel then follows its drift with a stochastic overlay and occasional multi-month shocks. Gas drifts up 1.7% a year at a volatility of 0.28, and hydrogen drifts down 2.8% a year.
Carbon on the standard world starts at nothing and is interpolated along $4 a tonne at year 8, $22 at year 15, $65 at year 25, $130 at year 40 and $190 at year 60.
Build times are the plant’s own: 6 months for a wind turbine or a solar park, 24 months for a combined-cycle gas plant, 60 months for a small modular reactor.
Households buy their own kit at retail prices, not bulk ones: $2,000 per kWp of rooftop solar, $10,500 for a heat pump over a replacement gas boiler, $14,000 for an electric car over the petrol one it replaces.
Road fuel carries a flat duty of $42/MWh thermal on top of the wholesale crude price and the supplier markup, which is a large part of what makes an electric car worth buying.

Weather, wind and solar

Nothing in the weather is a recorded time series. Temperature, cloud and wind each follow the season through one cosine of the day of the year, and each carries an anomaly that decays back towards normal with a fixed persistence: 46 hours for wind, 30 for cloud, 70 for temperature. That is why weather here lasts for days instead of flickering between steps. Heavy cloud together with weak wind is read as a blocking high, which cuts the wind target to 32 per cent and in winter pulls temperature down by up to one standard deviation of its own anomaly, which is 2.5 K at low latitude and 6 K at high. A collapse in renewable supply and a peak in heat demand are then one weather system rather than two coincidences.

Wind becomes power through one curve. Below 3 m/s the machine makes nothing. From there output follows the cube of hub-height wind speed until it reaches plate, less 1 per cent for transformer and cable losses, and above 25 m/s storm control feathers it down to a stop at 30 m/s instead of tripping off a cliff. The speed at which a turbine first reaches its rating is not a tuning knob. It falls out of the rotor, as rating over swept area divided by half the air density and a rotor power coefficient of 0.45, cube rooted. The catalogued 6 MW turbine on its 162 m rotor rates at 10.18 m/s, and the 15 MW offshore machine on 236 m at 10.76 m/s.

Solar starts from geometry. Declination and hour angle give the solar elevation, and clear-sky irradiance is 1120 W/m2 times its sine after air mass attenuation, peaking at 776 W/m2 at noon on day 172 at mid latitude. Full overcast still passes 22 per cent of that. Output is then irradiance over 1000, times a ground factor running from 0.55 to 1.30, times a derate of 0.4 per cent per kelvin above 25 C, times a flat 0.96 for the rest of the system. Snow comes off separately, up to 70 per cent of a plant’s output at full cover.

Storms are watched rather than announced. About 5.7 arrive a year at mid latitude, each lasting 5 to 22 hours and holding the wind at 16 m/s or above, up to 32 m/s at full intensity, which is past the feathering band and so takes the turbines to nothing. A harsh world multiplies the hourly storm chance by 1.4. The alert cards that once announced storms, dark doldrums, cold waves and heat waves were deleted on purpose, because a state that lasts days belongs on an instrument, and at 400 times speed a card came back every few real seconds.

A capacity factor is the share of nameplate a machine averages over a year, so a turbine at 46 per cent delivers 46 per cent of what it would deliver running flat out every hour.
Latitude is drawn between 0.33 and 0.6 for a normal world and between 0.62 and 0.78 for a harsh one. The figures here are at 0.45, where the climate’s mean wind is 7.28 m/s.
At that latitude the 6 MW onshore turbine averages 33.8 per cent of nameplate on the poorest ground it accepts, which is site quality 0.35, 45.9 per cent on middling ground and 60.5 per cent on the best.
A solar park averages 10.4 per cent of nameplate on the poorest ground it accepts, site quality 0.3, and 17.4 per cent on the best.
Those planning figures come from integrating the power curve over a Rayleigh wind distribution in 48 bins, and from walking the sun across a normal year in 24 sample days of 24 hours each. Three simulated years of live weather return 46.6 per cent wind and 13.1 per cent solar at a middling site, against planning estimates of 45.9 and 13.4, so the estimate and the running model agree to within a point.
Site quality multiplies the wind speed rather than the output, from 0.62 to 1.34, which is why better ground is worth more than the quality number looks like it should be.
Each year is given wind, solar and river indexes drawn once and fixed by seed and year, wind between 0.75 and 1.25 of normal and solar between 0.82 and 1.18. The planning briefing quotes an estimate of that year’s index, widened by however much accuracy research has not yet bought.

Voltages, lines and congestion

Four voltage levels exist, defined in one table: 380 kV, 110 kV, 20 kV and 400 V. They exchange power only through a building that transforms between them, and a rated transformer is modelled as two busbars with a winding between them rather than as one node, so power that steps down to 20 kV and climbs back up pays for both banks. The ratings are sized to a colony whose peak is about 60 MW, which is why the numbers below are small.

A circuit’s rating is its catalogue plate, raised 8% and then a further 5% by conductor research, and scaled by condition: a damaged line carries 0.35 plus 0.65 times its health, so a third of the plate still stands at zero. Losses are quadratic in flow, and one map tile is 600 metres of route. How a flow splits between parallel paths is set by susceptance, kilovolts squared over ohms per kilometre, taken from the voltage and the conductor and never from the rating, so a 20 kV feeder beside a 110 kV circuit has roughly a twenty-seventh of its susceptance and carries almost none of the transit. Each electrical island is solved as a DC load flow, factorised by sparse Cholesky when the graph is rebuilt, with a conjugate gradient solver kept as the fallback.

Congestion is an element carrying more than its rating, or more than 1.28 times its rating while an emergency overload is authorised, which takes 0.006 of condition an hour off every circuit that is over. Storage charging gives way first, in up to three rounds with the flow solved again between them. What is left comes off the customers standing downstream: the trace leaves the overloaded line or winding, crosses each edge only in the direction the power is actually flowing, and never steps back through the element itself. A geometric cut would answer that the whole colony is behind it, because a meshed colony is one island. A conductor held above its rating heats at a rate set by the square of its loading, 41% over reaches its limit in about half an hour, and protection then opens it for 45 minutes at a cost of 0.04 condition and recloses it still half warm.

An interior tile of a route is a junction when another route at the same voltage stands there without running alongside through both neighbouring tiles, so two circuits sharing a corridor are tied where they meet and where they part, not at every tile between. The flow solver, the division of a tapped route and the map all read that one rule, which is why the drawn wire and the solved circuit agree. A route’s own ends are never junctions, and an end landing on a tile nothing else touches is reported as loose.

380 kV overhead line: 1800 MW at 0.042% loss per kilometre and 0.26 ohm per kilometre, standing in for a four-bundle 265/35 circuit.
110 kV line: 80 MW for a single circuit, 160 MW double, 320 MW quad, at 0.19%, 0.16% and 0.14% loss per kilometre.
20 kV feeder: 14 MW for one cable, 28 MW for two, 56 MW for four, at 1.04%, 0.88% and 0.80% loss per kilometre.
The catalogue gives the 20 kV feeder and the lightest 110 kV circuit the same AL/ST 120/20 conductor: 14 MW at 20 kV and 80 MW at 110 kV, which is the case for stepping up stated as one number.
Transformers: a town is founded with a 25 MW substation, the catalogue 110/20 kV set passes 40 MW, the main substation 100 MW, the 380/110 kV substation 300 MW and the grid node 600 MW.
A 20/0.4 kV kiosk has no graph throughput at all: it is rated at 6 MW against its own streets, within a reach of 3 tiles.

Storage and hydrogen

Storage is a ladder of duration. The Grid Battery is 40 MW across 120 MWh, three hours, and gives back 88 per cent of what it takes. The Long-Duration Battery is 25 MW across 250 MWh, ten hours, at 60 per cent, because flow chemistry spends a third of the input on pumps and shunt currents. Hydrogen holds a season and returns least of all. Every store leaks while it stands, from 0.8 per cent of a full lithium pack a day down to 0.02 per cent for a salt cavern, and the two batteries and the electrolyser fade with cycles and years to a floor of a third of nameplate.

The step is fifteen minutes. A surplus is offered down a charging queue the player orders, and what the stores will not take is curtailed. Outside a surplus they charge only in a quiet hour, measured against a reference that smooths net load over two days, and only up to that reference, so the draw fills a valley instead of digging a new peak. Discharging is a bid: the smoothed clearing price, times a dial at a default 1.05, scaled from 1.25 at the reserve line to 0.6 when full, plus variable O&M, plus wear priced as capital over warranted throughput. Wear is $18.33 a megawatt-hour for the lithium pack and $2.27 for the flow battery.

The Electrolyser Plant draws 60 MW and stores it as hydrogen at 72 per cent at full plate, and at its best 76.9 per cent at 30 per cent of plate, where the improving stack crosses the fixed parasitic draw. It keeps a quarter of its 600 MWh buffer on site and pipes the rest to the vessels, smallest first, so a 900 MWh tank farm fills before a 12,000 MWh cavern. The Hydrogen Turbine burns it back at 40 per cent, which puts the whole electricity to hydrogen to electricity path at 28.8 per cent. Stored hydrogen is priced as an opportunity cost rather than at what it cost to make: the market price of hydrogen, $120 a megawatt-hour thermal in the founding year, scaled by the same fullness curve the batteries use. Settlement then pays only for what the vessels could not cover.

Storage can sit idle while wind is being curtailed. That was reported as a bug, measured over 120 simulated days at quarter-hour resolution, and found correct. Two causes account for it, and the probe treats anything else as a fault.

An electrolyser has a floor under it: 15 per cent of 60 MW, so 9 MW or nothing, because at low current density hydrogen diffuses through the diaphragm towards the flammability limit. Offered less than 9 MW the stack stays down and the surplus is curtailed beside it, rather than the colony starting a gas set to hold the stack above its own minimum.
A full hydrogen chain pins the stack. The electrolyser keeps a quarter of its buffer on site and moves the rest into the vessels, so once the tanks and caverns are full the buffer stays full and there is nowhere for the next megawatt-hour to go.
The tank farm and the cavern have no voltage and take no terminal on the electrical graph. They are filled through a pipe from the electrolyser beside them and have nothing to inject, so running a line to one changes nothing, which a probe was written to confirm.

Plants, condition and end of life

Every buildable thing is one catalogue entry, and that entry is the physical description: electrical capacity in megawatts, electrical efficiency as electricity out per unit of fuel heat in, capital cost as an absolute figure per unit rather than per kilowatt, fixed operations and maintenance per year, variable operations and maintenance per MWh produced, months to build, and design life in years. The listed price is a reference that a learning curve brings down as more of that technology family is installed, with a floor at 25% of list. Fixed maintenance accrues hour by hour for as long as the asset stands, whether or not it generates, and a mothballed machine still pays 40% of it.

Condition is a single number that multiplies the machine’s output, and it carries both wear and damage. Age is held separately, as a ceiling condition can never rise above. An unmaintained machine loses 0.45 of that ceiling spread evenly over the days of its design life, so it stands at 55% on the day the life runs out, and the ceiling stops falling at 0.25. Damage is the gap below the ceiling, and that gap is the part a crew can put back. Because one figure carries both, readouts score condition against the ceiling its years allow rather than against a new machine. An overhaul restores to a ceiling that itself falls with age: 0.28 of it across the design life, a further 0.45 for each design life beyond that, and a floor of 0.35.

Failures come from a base annual rate for the kind of machine, the running hours since anyone last opened it up, and condition already lost. A fuelled plant sits at 1.37% a year in good order, a wind farm at 1.9%, solar at 0.5%. One service interval is 30,000 running hours, and a full interval multiplies the base rate by 3.9. A fuelled plant that trips falls to between 25% and 60% of its ceiling; anything else falls only to between 66% and 88%, because one gearbox out of twenty turbines is a derate rather than a station coming off the system. Repairs then heal at 0.013 of condition an hour for a fuelled plant, 0.009 for most other assets and 0.005 for a wind site, and nothing interrupts the player unless at least a megawatt is lost and the loss exceeds the firm headroom left to cover it.

Destroying something serves a three month notice. It leaves the running system on the click, generating nothing and carrying nothing, and the crews arrive ninety days later; scrap worth 12% of what that machine currently costs to build is paid only on removal, so calling the notice off inside the window costs nothing. A retirement booked on the planning screen takes the unit off in the month chosen instead. Rebuilding in place costs 65% of a new build and takes the full build time, with the old machine still running at its present condition until the crews finish. Completing it resets condition, ceiling, running hours and the age clock, while the asset’s identity survives: its name, every circuit landing on it, its town and the programmes bought on that town’s streets. A weather-driven site can also be repowered once seven tenths of its design life is gone, at 55% of a new build and 45% of the build time, with a floor of three months.

The Wood-Fired Steam Plant is 18 MW at 22% efficiency, costs $26,000,000, takes eight months to build and has a thirty year design life.
The Combined-Cycle Gas Plant is 100 MW at 59% efficiency, costs $75,000,000, takes twenty four months to build and has a thirty five year design life.
The Small Modular Reactor is 60 MW at 34% efficiency, costs $300,000,000, takes sixty months to build and has a sixty year design life.
The Wind Turbine is 6 MW on a 162 metre rotor, costs $9,400,000, takes six months to build and has a twenty five year design life.
The Hydroelectric Dam is 25 MW with no fuel and a reservoir of a week at full output, costs $72,000,000, takes thirty months to build and has an eighty year design life.

What it does not model

This is the part that makes the rest of the page worth reading. Some of it was never built. Some of it was built, played, and taken out again because it cost more attention than it returned. Where a mechanic was removed, the figures it needed are still on the catalogue entries, so it can come back without the whole fleet being measured again.

Ramp rates and unit commitment are out of dispatch. Start times, start costs and minimum down times are catalogued for every thermal plant and read by nothing, so a machine reaches whatever it is asked for inside one interval and a restart is free.
There is no interconnector and no neighbour to trade with. A shortfall becomes unserved energy, and a surplus is spilled into dump banks and billed.
The network carries real power only. It is a DC load flow, so there is no reactive power, no power factor, no voltage magnitude and no tap changer in it. Frequency is a proxy that only a shortfall moves.
Turbines do not stand in each other’s wind. There are no wake or array losses, so ten machines in a row make ten times what one makes.
Solar runs on horizontal irradiance alone. Panel tilt, azimuth and tracking are one ground multiplier rather than a geometry.
Hydrogen is a single colony-wide stock measured in megawatt-hours thermal. There are no kilograms, no pressures, no compressors as separate machines and no pipe network.
Every route is an overhead line. Buried cable was in the catalogue and was taken out, and so was pumped storage, which had left the battery beside it with nothing of its own to say.
A breakdown is one condition number falling, not a named component failing. There is no parts inventory, no spares lead time and no crew roster.
There is no corporation tax and no general inflation. Apart from the fuel drifts, the carbon path and the technology learning curves, every price is in the same dollars for sixty years.
Capital costs are ranked rather than measured. The efficiencies, fuel prices, carbon intensities and minimum loads that a bid is made of are the real present-day figures, but the absolute dollars per kilowatt sit roughly a third to a half under the real ones. What the catalogue asserts is the order the technologies come in, not the price of a station.