Developer reverse engineers SimCity's scoring formula and breeds a near perfect city

A blog post picks apart the original SimCity's hidden scoring maths, using an open source port to breed cities that nearly max out the game's 1000 point scale.

AI-drafted from 1 source. This story was written with AI from the outside reporting listed under Sources, and checked by automated tools before it went up. The originals have the full detail.

A developer has spent time working out exactly what the original SimCity's mysterious “Score” number actually measures, then used that knowledge to breed cities that come close to a perfect 1000. The result, written up in a post titled A Perfect SimCity, is less a strategy guide and more an autopsy of a 1989 simulation engine, and it turns out the game rewards behaviour that no sane mayor would choose.

The author works from Micropolis and MicropolisCore, open source ports of the original SimCity ruleset maintained by Don Hopkins, and drives them through Python shims from a project called gym-city to run large batches of automated experiments. Ten resulting cities are shown off: five tuned purely for score, five tuned purely for population, the largest reaching 709,405 residents.

What actually scores well

The high scoring cities are strange places. Population sits at 12,000 to 16,000, tax is pinned at 1 percent, and roads and rail are sparse and deliberately disconnected. There is no fire department, and most of the top cities skip the police station too. That is because residential and industrial zones only check their twelve surrounding tiles for any road or rail link, and commercial zones just need a delivery vehicle to walk up to 30 steps from the perimeter, including the zone's own tiles. A short loop of track satisfies the requirement entirely. Rail carries no pollution and registers no traffic at all, while roads do both, and pollution suppresses residential growth. Police and fire departments are large budget items that must be funded at 100 percent or the score suffers badly, so the cheapest route to a high score is simply not to build them, and to keep tax as low as the budget allows. Raise tax to 2 percent and the score drops by nearly 30 points on its own.

The unemployment loophole

The scoring model runs in two stages. The first stage subtracts seven penalty terms, for crime, pollution, land value, tax rate, traffic, unemployment and active fires. Six of those are bounded at zero, meaning they can only ever cost points. Unemployment is not. The post lays out the formula, roughly resident population divided by eight times available jobs, minus one, times 255, and when jobs outnumber zoned residents the term goes negative and adds points back, worth around 128 in the cities studied. No other factor moves the score anywhere near that much, which makes an oversupply of jobs the single most powerful lever in the whole system.

A second stage applies nine further penalties on top, each one shrinking whatever is left. The first three punish a city for failing to build a stadium past 500 residents, an airport past 100 commercial, and a seaport past 70 industrial. A stadium costs nothing in traffic or pollution and can go anywhere, so the algorithm happily builds one. Airports and seaports are dense polluters, so the highest scoring cities skip them and simply eat the penalty instead.

A fountain that counts as a slum

The oddest finding concerns land value scanning. The game decides what counts towards each averaged scoring term by tile ID, and grass, forest and park tiles under ID 64 are invisible to the calculation. A park fountain, tile 840, is not. It gets counted even though nobody lives there, and because it contributes no crime of its own, a grid of fountains can drag the citywide crime average down by 33 points without a single resident being safer. It is the kind of quirk that only shows up once someone reads the scoring code line by line rather than just playing the game.

None of this makes for a city anyone would actually want to live in, empty of services, starved of housing density, dotted with ornamental fountains for their statistical effect rather than their charm. But for anyone building a simulation game today, it is a useful warning about the gap between the numbers a scoring system can measure and the outcomes a designer actually wants to reward.

Sources

• A Perfect SimCity (Lobsters games)

Image: Illustration: AI-generated for Gaming World