Make the economy actually run: close the loop, fix pricing and trade

[abogun-product]

The four layers of the simulation were all sketched out, but none of them were connected, so nothing could actually happen. This makes the whole loop run, and adds the instrumentation needed to see whether the economy converges or quietly explodes.

What was broken

• Trade did not exist. FindBestDeal was never called from anywhere, and nothing ever wrote to tradeStorage, so City.Trade() was a no-op. No code moved goods or money between cities.
• The arbitrage maths mixed up its units. amount multiplied an amount of the buyer's resource by a rate that had to divide it, and the two legs of the round trip used inverted barter rates, so the profit was neither in the right units nor of a reliable sign.
• Pricing destroyed itself on turn one. CalculatePrice ran first in the turn, before demand and offer were filled in, so max(0,1)/max(0,1) overwrote every base price with 1. All relative prices collapsed to parity and there was nothing left to arbitrage.
• The loop never closed. productionAmount += accumulated forever and nothing was ever consumed, storage only mirrored that running total, Pop.coins was never assigned so wealth was always 0, and supply grew linearly without bound.
• netcoreapp3.1 has been out of support since December 2022.

What this PR does

1. Builds and runs again. net9.0, nullable and implicit usings on, warnings as errors, dead usings gone, solution moved to the repository root, a test project added.

2. Closes the economic loop. Stocks and flows are separated, pops own their goods and their money, they buy what they cannot produce, eat every turn, and leftovers spoil. Coins stop being a resource: the crafter now produces tools, and money is the numeraire every price is quoted in. The money stock of the world is closed and never changes — there is a test for it.

3. Rewrites price formation. A price is a stock nudged towards the demand/supply ratio by at most MaxPriceStep per turn and clamped to a range. Two details turned out to matter far more than they look:

• Demand counts only what buyers can actually pay for. Counting the wishes of a bankrupt city keeps pushing its prices up, which is the last thing its broke citizens need — with gross demand the whole economy seizes up inside a hundred turns.
• Demand is elastic: cheap goods are worth stockpiling, dear ones are done without. Otherwise demand is a hard ceiling and supply a hard floor, any lasting imbalance compounds at 10% a turn until it hits a clamp, and a pop that has accumulated money has nothing left to spend it on, so the money never returns to the producers.

4. Makes arbitrage real. Deal is now one resource and the gap between two markets, sized against every constraint that applies — trade power left this turn, the share of the local surplus that may be stripped, the traders' capital, and the demand the destination can pay for — and it executes: goods and money change hands. Traders work in both directions, exporting a local surplus and importing what a neighbour sells cheaply. Transport costs a share of the cargo rather than a fee in money, because a money fee burns the fixed money stock turn after turn until everyone is broke (it did: 3300 of 4000 money gone in 200 turns).

The turn order is now produce -> price -> trade -> local market -> consume -> spoil. Trade has to run before the local market: the other way round the locals spend their money on the insufficient local supply first, the imports have no buyer left, and the traders go bankrupt and stop trading for good.

5. Makes a run observable. Deterministic seed, a per-turn CSV time series in long format, and every tunable number in SimulationConfig, overridable from the command line so balancing does not need a rebuild:

dotnet run --project TradeCraftSimulation -- --turns 300 --quiet --csv run.csv --config TransportLossShare=0.02

Does it work?

After 300 turns, prices settle into stable bands with no runaway and no collapse, and the result the model was built to produce shows up on its own — every good ends up cheapest in the city that specialises in it, on every seed tested:

seed	food cheapest in	wood cheapest in	tools cheapest in
1	Farmland (0.49)	Forresty (0.14)	Craftovo (1.55)
7	Farmland (0.49)	Forresty (0.15)	Craftovo (1.56)
42	Farmland (0.49)	Forresty (0.14)	Craftovo (1.41)
99	Farmland (0.52)	Forresty (0.15)	Craftovo (1.64)

The capital works as the entrepot the provinces trade through, satisfaction settles around 0.8, and trade is still running at turn 500.

Tests

42 tests, dotnet test. They cover the invariants that are easy to break and hard to notice: money and goods conservation, no pop ever spending money it does not have, prices staying inside their range, the same seed replaying the same run, stocks not running away — and that trade does not quietly die out mid-run, which is the failure mode that hid behind every other bug here.

Balancing decisions worth arguing about

These changed behaviour rather than fixing a bug, and are the parts most worth a second opinion:

• The merchant class had to shrink (TraderShare = 0.3, so the capital has 6 traders rather than 20). Arbitrageurs compete away the very margin they live on, so a merchant class of 35 in a world of 200 people cannot be fed by the margin on ~200 units of trade per turn. With the original counts the traders went bankrupt by turn 30 every single time, and trade stopped for good.
• The capital gets a 1.25x output bonus on everything. A quarter of its people produce nothing, so without it the capital is permanently short of every good, pays for the imports out of a fixed money stock and goes broke.
• Per-capita needs are calibrated to total world output over total population, so the world is exactly self-sufficient while every individual city is short of something. That gap is what makes trade worth doing.

Known limitations (documented in the README)

Nobody changes trade, so a glutted trade stays glutted: a woodcutter in a city drowning in wood is poor forever instead of becoming a farmer, and a bankrupt merchant stays bankrupt. Labour mobility is the single biggest thing still missing, and demography is the second.

[abogun-product]

Closing: moving this work to a private repository under abogun-product instead of a fork.