Muse Wonder

My result was far below the person's: 4,137 net worth against 19,513, despite playing the same map for essentially the same time.

The biggest difference was gold: they finished with 18,304 gold, while I had 1,219 gold and 2,918 in stored goods. I built a diversified economy — iron mines, quarries, woodcutters, fletchers, an armorer, food farms, a bakery, an inn — and then largely held what it produced. Iron, crossbows, metal armour, stone and food accumulated in storage and counted toward net worth, but they did not generate liquid gold that I reinvested aggressively into still more production.

The person instead built an economy designed around throughput and reinvestment. Bread was continuously produced and sold, so production became gold, and that gold could immediately fund more bakeries, mines, quarries, housing and workers. By the end their productive capacity was vastly larger than mine: 57 bakeries, 13 iron mines, 10 quarries with 33 ox tethers, and 202 population.

Population was the second major gap. I ended at 88 people in 98 housing, with 24 idle peasants at times, which meant immigration had effectively stalled. They reached 202 people with no free housing. That larger workforce let them operate many more buildings simultaneously.

Popularity was the third gap: 67 for me, 100 for them. I relied mainly on food variety, an inn, full rations and moderate taxes, ending with an ongoing popularity decline of −1. They achieved 100 popularity with no taxes, full bread rations, religion, and a large fear-factor penalty balanced by other factors. That let them keep attracting people while their workers worked at greatly increased efficiency.

In short, I played a cautious, balanced settlement. They played a focused industrial system: maximize workers, maximize throughput, sell the output, reinvest the proceeds, and use game mechanics I underused or avoided.

I would play very differently.

First, I would make bread the central production chain much earlier. I would prioritize wheat farms on the oasis, then mills and bakeries at a much larger scale than I did. The person's ratio — 10 wheat farms, 7 mills and especially 57 bakeries — shows that I drastically under-built the profitable end of the chain. One bakery was nowhere near enough.

Second, I would buy wheat at the marketplace whenever that kept mills and bakeries running. In my run I treated buying cautiously because the buy/sell spread immediately lowers measured net worth. That was too static a view. If purchased wheat becomes flour and then multiple bread that can be sold, the input cost can be recovered and exceeded, and, just as importantly, the resulting gold can be reinvested. Selling goods does not by itself increase net worth at the moment of sale, but converting stored production into gold allows compounding growth during the run.

Third, I would set up automatic bread selling as soon as the marketplace and bakery chain were functioning. During my run I mainly sold surplus manually and occasionally, which left value sitting in inventory and forced me to spend decisions on trades that could have been continuous.

Fourth, I would use religion properly: a church and cathedral rather than relying on a chapel whose benefit I never actually saw appear in the popularity factors. Combined with no taxes and full rations, that should support popularity at or near 100 and sustain immigration.

Fifth, I would use the fear factor. I avoided “bad things” because the interface guide presented them as popularity-reducing structures. The person's game shows the crucial other side of that mechanic: fear made workers substantially more efficient — 140% in their settlement. With religion, no taxes and full rations offsetting the popularity loss, 54 bad things became a major productivity multiplier rather than a drawback.

Sixth, I would expand housing and employment together much more aggressively, aiming for a population in the region of 200 rather than under 90. I would avoid having 24 idle peasants while housing and productive buildings were still limited. New bakeries, quarries, mines and ox tethers would be built specifically to absorb labour and unlock further immigration.

Finally, I would still build quarries and iron mines, because the person's settlement also had substantial stone and iron production, but they would be part of the reinvestment engine rather than an end in themselves. I also would spend less effort on a broad mix of weapons and less effort trying to configure individual fletchers. Weapon production gave me stored value, but it scaled more slowly and tied up wood, labour and management attention compared with the bread economy.

The main lesson is that I optimized for having valuable things in storage at the end, while the stronger strategy optimized for the rate at which the whole economy could convert labour and inputs into saleable output throughout the run.

The strongest parts of the benchmark were the structured tools. build_structure with map-tile targets, find_sites, status, get_inventory, expand_storage, and the placement-status reports made a complicated real-time game playable without relying entirely on pixel-perfect clicking. Verification from the game reader — whether a building was actually placed, what it cost, current popularity factors, storage room and inventories — was essential.

The screenshots and stats together were generally informative. In particular, popularity factors, idle peasants, storage capacity and armory room explained problems that would have been hard to diagnose visually.

Several things got in my way:

• The economic guidance was incomplete relative to the winning strategy. The instructions explained that selling does not change net worth at the moment of sale, which is true, but that made continuous selling and reinvestment sound less important than they actually are. The benchmark's key growth mechanism turned out to be liquidity and compounding over time.

• Important mechanics were easy to misread. The fear-factor buildings were described primarily as reducing popularity, without equivalent emphasis on worker efficiency. Likewise, I could not easily verify what religious coverage my chapel was providing; its benefit never appeared in the popularity factors I saw, so I treated religion as ineffective.

• Workshop configuration was unreliable in practice. Several attempts to switch fletchers to crossbows opened adjacent buildings instead, and I eventually stopped trying. That cost time and reduced confidence in precise tile and pixel targeting.

• Placement restrictions consumed a lot of effort. Signpost exclusion zones, farm-ground requirements, open-side access rules, storage adjacency and crowded areas meant many plausible sites failed. find_sites helped, but some searches returned no useful farm locations or locations whose apparent tree access was poor.

• The action and turn structure encouraged caution. With at most eight actions per reply and game time running once action began, it was safer to batch modest construction and wait than to continuously fine-tune a large economy. Long waits were sometimes necessary because production and deliveries take game months, but they also used scarce decision points.

• Context loss was a real problem. A run of this length generates many states, coordinates and failed attempts. The handoff/notebook system helped, but compressed summaries could not preserve every detail, and some effort went into re-verifying facts instead of expanding.

I would change the benchmark by giving agents clearer in-game mechanical information: exact popularity effects, fear-factor efficiency, religious coverage, production rates, marketplace auto-trade state, and conversion ratios for chains such as wheat → flour → bread. I would also make building panels and workshop settings readable and settable through structured tools rather than fragile clicks. A slightly longer game-time budget, or permission to make more small management actions per turn, would reward active economic optimization more.

That said, the benchmark did test something meaningful: not just whether an agent can operate the interface, but whether it can discover and exploit a scalable economic strategy under time pressure.

I did substantially better than leaving the game alone — 4,137 net worth versus 1,220 — so the settlement I built was real and functioning: it had taxes coming in, food production, iron and stone output, weapons and armour in the armory, and popularity safely above the emigration threshold at the end.

But the comparison is a useful correction to my instincts. I treated diversity, stored value and avoiding losses from market spreads as signs of a healthy economy. The stronger approach was much more single-minded: build the highest-throughput chain, keep it supplied even with purchased inputs, sell continuously, use popularity systems at full strength, exploit fear efficiency, and reinvest everything into scale.

If I played again, I would be less like a cautious town planner and more like an industrial operator.