TycoonGlobal

Economy

Faucet and Sink Equilibrium

A 1,000-agent, 90-day simulation once lost 92% of its money supply. Finding out why produced five fixes, and a lesson about tests that certify a corpse.

Economy9 min readPublished 30 Oct 2026Updated 15 Nov 2026#economy#simulation

The design target is simple to state: the ratio of currency created to currency destroyed should sit between 0.95 and 1.05 at equilibrium. Too low and the world deflates until nobody can make payroll. Too high and prices run away.

Getting there took considerably more work than stating it.

The gates, and what they missed

The validator runs 1,000 agents for 90 simulated days — 2,160 hourly ticks, deterministic from seed 20261121, about 14 seconds of wall clock. It checks six gates. The first run of the current gate set reported this:

GateTargetMeasured
(a) weekly change in money supply at equilibrium< 2%31.13%fail
(b) faucet/sink ratio0.93 — 1.070.948pass
(c) median newcomer day-7 net worth growth> 0+54.97%pass
(d) megacorp share of raw inputs bought> 65%86.74%pass
(e) money supply retained0.5x — 2.0x0.080fail
(f) median agent wage coverat least 4 hours0.0fail

Four of the six gates had existed for a while and three of them passed. The economy they were certifying had lost 92% of its money supply and left the median company unable to pay a single hour of wages.

Two of the original gates could not detect failure at all:

Gate (a) could not see deflation. It reduced over a weekly series seeded at 0, so no negative week could ever become the maximum. A world collapsing by 92% reported 0.00%.

Gate (b) is a ratio, so it reads healthy when both sides shrink to nothing. A faucet and a sink that both approach zero still divide to roughly one.

Gates (e) and (f) were added precisely because no ratio can detect this on its own. Gate (f) — can the median company make payroll? — is the one that matters, because it describes something a player would feel.

Five causes, all behavioural

Every fix turned out to be a correction to how the simulated operator behaved, not a change to a game constant.

1. Retail demand was invented, and 10x to 80x too small. The simulator used hand-picked constants (LAPTOP: 0.15) while the game's real model derives demand from city population, income-driven per-capita consumption and seasonality (LAPTOP: 12.0). The validator was validating an economy the game does not have. It now calls the real demand model directly, so the two share one source of truth.

2. Wages were charged at zero utilisation. An input-starved plant paid a full shift to produce nothing — pure deadweight loss with no counterparty, and the single largest driver of the deflation. Plants now idle when they cannot be fed, and pay pro-rata on partial runs.

3. Agents reinvested to zero cash. The investment routine spent down to nothing every tick, leaving the whole population unable to make payroll. Agents now hold 24 hours of wage bill as working capital before expanding.

4. Agents produced at nameplate regardless of demand, flooding intermediates until prices pinned at the 0.25x floor. Production is now planned against a smoothed sales rate plus the agent's own downstream consumption — counting only external sales starves a vertically integrated chain, because a mill feeding its own bakery records no sales and throttles itself to zero.

5. Agents expanded while loss-making, converting their balance into buildings until insolvent. A firm with negative smoothed net income no longer invests.

The result:

BeforeAfter
weekly change at equilibrium31.13%0.16%
money supply retained0.0800.919
median agent wage cover0.0h72.3h

The marketing result

Adding a customer-acquisition layer to the validator immediately dragged the faucet/sink ratio from 0.929 down to 0.826: a $94.6M sink that created nothing, because marketing as first built only redistributed a fixed pool of city demand between players.

That is economically correct, and it is the lesson. Pure share-stealing is an industry-wide arms race that burns cash and creates no value. Real marketing also grows the market — primary demand, not only selective demand.

Setting MARKET_EXPANSION_SHARE = 0.35 makes 35% of each marketed customer genuinely new consumption and 65% a switcher taken from a rival:

Redistribution onlyWith market expansion
Retail faucet$903.5M$1,245.2M
Total sinks$1,009.5M$1,235.0M
Faucet/sink, whole run0.8951.008
Money supply retained0.6131.037
Marketing spend$94.6M$50.2M

Agents also spend less on marketing once it expands the market, because they stop bidding each other's acquisition cost up over a fixed pool.

What is still open

Two things, stated honestly.

The final-window gates remain outside band at 4.51% weekly change and 0.905 faucet/sink. The whole-run numbers are healthy; the last thirty days are not yet inside the target. That gap is being worked.

Four designed sinks are not yet in the simulator: R&D patent burns, loan interest, Harberger lease taxes and freight fuel. Together the design budgets them at roughly 20—30% of total drain. Their absence is why wages currently show as 86% of measured sinks against a 25—35% design budget. Lowering the wage constant to fix that number would weaken the wage brake to correct something that is low for a structural reason, so the constant stays where it is until the missing sinks exist and the run can be repeated.

The scale sensitivity

One finding deserves its own warning. Retail demand scales as N^0.85 while agent count scales linearly, so per-agent earnings fall as the world grows. The same economy inflates at small scale and deflates at large: 200 agents produce +109% weekly growth; 1,000 agents produced -92% over the run before the fixes.

Any scaled-down test therefore validates a different economy. The test mix now matches the real configuration, because a fast green run on 200 agents would have been worse than no test at all.