The Capacity Economy (as shipped)

Under the live market mode (marketSystemMode: "plants", the top tier of src/lib/market/modes.ts10d ago), a corporation's sector no longer earns from a self-compounding revenue number. It owns physical capacity (capitalStock, measured in output units per day), buys more of it through a priced build queue, staffs it with workers, and earns only what those plants actually produce and sell into a clearing market. This document describes that system as it runs today, from the player-visible loop down to the constants and formulas in src/lib/constants/capacityEconomy.ts6d ago, src/lib/market/capital.ts10d ago, and src/lib/turn/corporation/sectorTurn.ts1d ago. It supersedes the older market-mode sections of design/corporations.md and design/economic-systems.md where they describe growth-slider revenue as current behavior.

The player-visible loop#

1. Found or expand. A CEO opens a sector in a (state, sectorType) cell. Founding builds are priced at CAPACITY_FOUNDING_DISCOUNT = 0.1 of the standing capacity price.
2. Build capacity. POST .../build-capacity (src/lib/corporations/commands/sectorOperations/buildCapacity.ts7d ago) charges cash now; the order sits in the sector's buildQueue and comes online CAPACITY_BUILD_TURNS(type) turns later (12 turns for retail up to 96 for energy and extraction; default 48; 48 turns = one game year). Cancelling refunds CAPACITY_BUILD_CANCEL_REFUND = 0.75 of the paid cost. A sector holds at most 20 outstanding orders; a single order is capped at MAX_BUILD_UNITS_PER_ORDER = 10,000,000 units.
3. Run the plants. Each turn the sector produces up to its capacity, throttled by inputs, disasters, strikes, and its production-policy slider; it sells into the clearing market at its chosen pricing posture (clampPricingPosture, −20% to +20%); revenue is derived from units sold, not asserted.
4. Pay real costs. Labour (workers × wages), physical input purchases, idle upkeep, compliance, and the residual opex line. Idle capacity is not free: cost basis is utilization + IDLE_UPKEEP_FRACTION × (1 − utilization) with IDLE_UPKEEP_FRACTION = 0.3. Mothballing (action: "mothball") cold-stows the plants: zero output, upkeep at MOTHBALL_UPKEEP_FRACTION = 0.2 of running maintenance.
5. Grow, shrink, retool, or fight. Depreciation (CAPITAL_DEPRECIATION_PER_TURN = 0.0005, about 12% over five game years) erodes capacity that is not replaced. Retooling to another production strategy rescales the stock so the nameplate is invariant (D9, below). Corporate attacks transfer capacity at ATTACK_CAPTURE_EFFICIENCY = 0.6 (the rest is destroyed) and are floored at the build price of the received units times ATTACK_BUILD_PRICE_PREMIUM = 1.15, so war is always dearer than building.

The interface counts in facilities, not raw units. src/lib/constants/facilityQuantum.ts10d ago sizes one facility per sector so its nameplate lands near FACILITY_TARGET_DAILY_REVENUE_ANCHOR = ₳25,000/day (energy: 250 units per power station; automobiles: 1 unit per assembly plant; retail: 80 per store). src/lib/constants/facilityVocabulary.ts10d ago supplies the display nouns and verbs: you open a store, build an arsenal, sink a mine. Neither file changes a number; the engine, storage, and market share stay in units.

The unit of capacity and RPU#

One unit of capacity is "one output unit per day". Its money value hangs on one quantity, the unit yield

k(type) = Σ over output commodities (rate_c / basePrice_c)   [units per ₳ of daily revenue]
RPU(type) = 1 / k(type)                                       [₳ of daily revenue per unit]

exactly the per-revenue slope of impliedOutputUnits in src/lib/market/capital.ts10d ago. RPU is a harmonic-style mix price dominated by the cheap legs of the output basket, which is why an energy unit is worth about ₳92/day while an automobile unit is worth about ₳50,000/day. plantsMixPrice in sectorTurn.ts is the same number, so the stored pair always satisfies revenue === capitalStock × mixPrice exactly (capacity is persisted unrounded for this reason).

Two derived anchors, both scale-free and both pinned by tests against the legacy tables:

• Identity B (price). capacityPricePerUnit(type, year, unitScale) = GROWTH_COST_MULTIPLIER × RPU × capacityEraPriceIndex(year), with GROWTH_COST_MULTIPLIER = 3.0 (src/lib/constants/corporations.ts14h ago). At the 1953 calibration anchor the era index is exactly 1.0, so building through the queue costs exactly what the legacy growth slider charged for the same increment.
• Identity A (labour). laborIntensity(type, year, unitScale) = RPU / CAPACITY_REVENUE_PER_WORKER × capacityEraLaborIndex(year), with CAPACITY_REVENUE_PER_WORKER = ₳20 (mirroring the module-private REVENUE_PER_WORKER behind calculateWorkers, also ₳20). At 1953 the labour index is 1.0, so capacity staffed from this table carries exactly the headcount calculateWorkers gives it.

How a build is priced#

computeBuildCost is the single pure function the command, the UI preview, NPP behavior, and the tests all share:

total = units × capacityPricePerUnit(type, year, eraUnitScale)
              × dominanceMult(share)          // local vs national share, harsher leg wins
              × rateMult(primeRate, acumen)   // max(0.5, 1 + primeRate/10 × acumenRateSensitivity)
              × acumenMult(acumen)            // flat CEO Business Acumen discount
              × techMult                      // capped tech growth-cost reduction, clamped (0, 1]
              × hostPriceMult(costOfLiving)   // host state's index / 100, clamped [0.6, 1.6]
              × foundingMult                  // 0.1 on the founding build, else 1

The host multiplier reads the state's costOfLiving metric (a real index, deliberately not an exchange rate, which would create a weak-currency carry trade); the clamp band is HOST_BUILD_PRICE_INDEX_MIN = 0.6 to HOST_BUILD_PRICE_INDEX_MAX = 1.6. Under plants, dominance is tolled here and only here: the old permanent dominance margin penalty and regulatory-burden revenue tax are faded out over the plants ramp in sectorTurn.ts, so market leadership is a barrier to expansion, not a tax on operating.

Era money scale#

The base-price tables are 2019-calibrated. Each world carries an era unit scale, getEraUnitScale(preset) in src/lib/constants/sectorSeedEra.ts10d ago (1 for modern worlds, roughly 70 for 1953), threaded as a required parameter through every ₳-to-units conversion so a 1953 economy does not collapse into a handful of modern-sized units. The era price column (CAPACITY_ERA_PRICE_SPANS) steps 1.0 (through 1970), 1.4 (1971 to 1978), 2.6 (1979 to 1990), 3.6 (1991 to the modern boundary), and 5.0 modern; the labour column derives live from eraLaborMultiplier renormalized to 1.0 at 1953. Facility sizes are era-invariant by construction: RPU and "meaningful revenue" shrink by the same nominal ratio.

The sector turn#

processSector in src/lib/turn/corporation/sectorTurn.ts1d ago runs each sector once per turn inside the corporationTurn phase:

1. Deliver builds, advance capacity. Orders with onlineTurn <= currentTurn convert into stock (smooth orders deliver a slice per turn); their cash moves from construction-in-progress into capacityBookAnchor, the depreciated paid basis. Capacity then advances by advanceCapitalStock: stock × (1 + growth − CAPITAL_DEPRECIATION_PER_TURN), with growth pinned at 0 under plants (the slider no longer builds capacity). Queue writes are $pull/$inc deltas, never whole-array $set, so an order placed mid-turn survives.
2. Produce. producedUnits = capacity × productionFactor, where the production factor multiplies the disaster physical leg, the production-policy slider, nationalization transition shock, the extraction geological hard ceiling (plantsExtractionHardMin, the deposit is a second hard min on units with no 0.5 floor), input throughput (Leontief on lagged global balances), tech output multipliers, and the strike throttle (STRIKE_REVENUE_THROTTLE). A mothballed sector produces 0.
3. Sell and derive revenue. plantsDerivedHourlyRevenue = producedUnits × mixPrice / TURNS_PER_DAY × clearingRevenueLeg × embargoRevenueFactor. The clearing leg comes from src/lib/market/clearing.ts2d ago: sellers post a posture in [−0.2, +0.2], demand (lagged one turn) fills cheapest-first, and the leg factor is soldFraction × (1 + posture) × priceRealizationFactor. Quality scales only the premium portion of a positive posture (multiplier clamped [0.5, 1.5]).
4. Governor. The derived amount is blended against the pre-flip counterfactual baseline by softenedMarketRealizationAmount (src/lib/market/capital.ts10d ago): deviation is capped at MARKET_REALIZATION_DEVIATION_CAP = 0.15 of baseline and ramped in over MARKET_REALIZATION_RAMP_TURNS = 240 turns, and the cap widens as cap / (1 − λ) so it releases entirely at full ramp. λ = 0 on the flip turn makes enabling plants a byte-identical no-op; a fully ramped world stands on its physical result alone. The same λ (plantsRampLambda) fades in every other plants-only economics change (idle upkeep, dominance consolidation, the extraction hard min, the disaster physical leg).
5. Cost the plants physically. Under plants the margin-formula cost is replaced by physical lines (src/lib/corporations/physicalPnl.ts2d ago): inputs bought at lagged market prices for what was actually produced, labour, idle/mothball upkeep, compliance, financial legs, and a calibrated residual (otherOpexPerUnitAnchor, solved once on the first producing turn so the flip reproduces the old cost exactly, then held per unit). Profit is revenue minus these lines; the reported margin becomes an output (profit ÷ revenue), not an input.
6. Write back. sector.revenue is restated as the nameplate capitalStock × mixPrice (never realized revenue, which would compound the realization legs into the base at roughly −7%/turn); realizedRevenue, producedUnits, soldUnits, utilization, and clearing/throughput telemetry are persisted for display only.

Staffing and the labour market#

Headcount is calculateWorkers(revenue, workforceSkill): revenue / 20 at neutral skill 50, with a skill multiplier down to 0.70× at skill 100. When the labour system is on (labourSystemMode >= "wages"), a labour cost is carved out of maintenance as workers × wage-per-worker, scaled by the CEO wage slider, the minimum-wage Kaitz floor, tech automation, and the union premium; the split is profit-invariant at baseline. Unionization trends toward a condition-driven target each turn (wages vs cost of living, unemployment, union law, owned-union membership pressure), and strikes trigger with hysteresis and cooldowns (src/lib/labour/strikes.ts2d ago), throttling revenue and hitting margin (STRIKE_MARGIN_PENALTY_PP) while active. Per-state wage and automation indices are accumulated during the sector pass and feed the state labour metrics.

Market share and dominance#

src/lib/corporations/marketShare.ts13h ago: share = sector revenue ÷ effective market × 100, where the effective market is the larger of the GDP-derived floor (SECTOR_MARKET_GDP_FRACTION of state GDP split across sector types) and the owned-plus-unowned revenue sum, all in anchor ₳. Two dominance legs exist: the local (state, sectorType) share with a 50% threshold, and the national share (weighted average across states, 30% threshold); every toll charges the harsher leg. Under plants those tolls collapse into the build-price multiplier, as above.

Valuation#

sectorNPV is a perpetuity on current yearly profit (NPV_ANNUAL_DISCOUNT_RATE). Under capital and plants, valuation is floored by advanceCapitalBookAnchor: seeded at NPV on first exposure, ratcheting up with NPV, decaying at the depreciation rate when NPV falls, so a corp that owns real capacity through a transient profit dip is not valued as if it owns nothing. The paid basis (capacityBookAnchor) depreciates with the stock, so a half-worn plant books at half what was paid, never at half list price. Free capacity (grants, R&D) dilutes the per-unit basis deliberately: it cannot be exited for cash it never cost.

Retooling (D9)#

Capacity units are not commensurable across production strategies (a coal mix prices around ₳60/unit, rare earths in the tens of thousands). On a strategy change, rescaleCapacityForStrategyChange multiplies the stock (and in-flight unitsOrdered) by RPU_old / RPU_new so the nameplate capacity × RPU is invariant: a retool is a re-aim, not a capital grant. The rescale happens once at commit against the final rates; the 12-turn transition blend (STRATEGY_TRANSITION_TURNS) misprices the nameplate briefly, decaying to zero, which is accepted for auditability.