One episode · Oasis by the Sea

Muse Wonder

Benchmark 1.5.1MetaReasoning mediumPlayed by the maintainersFull-budget score

All of Muse Wonder’s results and comparisons

Score

51.4

Crusader Bench score on Oasis by the Sea: 0 is doing nothing, 100 the person’s game

Net worth

9,603

+8,383 over doing nothing, 46% of the person’s gain of 18,293

Population

103

50% of the person’s gain of 198 · popularity 97

Cost

$2.49

5.2M tokens

Time

33 game min

47 min of wall-clock time, 106 turns

Minute by Minute

Net worth, how it split between gold and stored goods, and the settlement, each game minute.

Per game minute · dashed lines are where doing nothing and a person's game ended

Muse Wonder
Baseline (do nothing), final
A person's game, final

Net Worth

gold plus stored goods at the sell price

Gold or Goods

Gold
Goods

Population

peasants

Popularity

0–100

Food in the Granary

units

Crusader Bench · crusbench.com

Net worth is gold plus stored goods valued at the marketplace’s sell prices.

Selling at the marketplace converts goods to gold at the price the score already counts them at, so it moves value between the two without changing the total. Buying costs net worth.

Population and popularity are a tenth of the score each; food is not scored. Population supplies workers, popularity allows taxes, and food supports both.

The episode

Watch the Episode

Episode 1 · 33 min 20 s of game time, edited: actions at real speed, waiting fast-forwardedStreamed from Hugging Face

ScorecardFull-budget score

9,603net worth

Gold: 787
Stored goods, at sell price: 8,816
Over the baseline (do nothing): +8,383
Population / housing: 103 / 146
Popularity: 97
Food: 126
Structures (map-wide): 116
Attempt: 1

The settlement at the endcentred on the keep, zoomed out

No final overview was published for this episode.

Final stock8,816 gold in goods

Good	Units	Price	Value
Gold	787	1	787
Iron	225	23	5,175
Stone	289	7	2,023
Flour	78	10	780
Bread	126	4	504
Wood	262	1	262
Wheat	9	8	72
Net worth			9,603

Beside a person’s gameOasis by the Sea, the same game time

	This episode	The person	Left alone
Net worth	9,603	19,513	1,220
Gold	787	18,304
Stored goods, at the sell price	8,816	1,209
Population	103	202
Housing	146	202
Popularity	97	100

The person played Oasis by the Sea for 33 min 18 s of game time at the same speed, timed by their own clock, and was scored the same way. The model saw this comparison only in its interview.

The town’s booksread from the game at the end

Date in the game: October 1200
Idle peasants: 24
Popularity next month: +2
Food: +4
Taxes: −2
Taxes: Low taxes, 61 gold a month
Rations: extra; 1 of 1 foods eaten
Worker efficiency: 100%
Storage: 28 stockpile piles, 6 empty; 1 granary holding 126 food

Buildings83

Hovel17
Iron Mine12
Woodcutter11
Bakery9
Ox Tether9
Stockpile7
Quarry5
Wheat Farm3
Apple Orchard2
Dairy Farm2
Mill2
Chapel1
Granary1
Keep1
Marketplace1

Tool calls183

find_sites48
build_structure29
market_trade25
building_info22
go_to_tile21
wait_and_observe13
center_on5
place_near5
set_tax5
map_overview3
expand_storage2
inspect_building1
notebook_write1
set_rations1
status1
update_plan1

Play

Turns: 106
Buildings placed / attempted: 70 / 75
Anchor-tool placements: 9
Tool errors: 0
Raw inputs sent to the game: 593
Context compactions: 2
Game time: 33 min

Compute

Tokens: 5.2M
Served from cache: 96%
Output tokens: 38K
Cost, as billed: $2.49
Wall-clock time: 47 min
of which inference: 18 min
Peak game memory: 4,429 MiB

The plan, before play

Task understanding

Play the freshly loaded “Oasis by the Sea-1” Free Build map for 33.3 minutes of game time.
Final score is based on the settlement left standing at the end: net worth (gold + stored goods at marketplace sell prices), population, and popularity. I must balance all three.
Starting package: ~1,000 gold, 50 wood, 25 stone, 50 bread (arriving over the first ~25 game seconds), a keep and stockpile already placed, housing for 10.
No soldiers, walls, towers, gates or other defences — they waste resources for this benchmark.
High-value growth comes from production: stone (7 each) and especially iron (23 each), food surplus (4 each, flour/ale 10), plus steady tax gold once popularity supports it. Buying goods loses the buy/sell spread, so I’ll only buy early wood/materials if it speeds up productive buildings enough to repay it.
Key constraints: farms only on scarce oasis grass SW of the keep; woodcutters near the western tree grove; iron ore just north of the keep, stone ~50 tiles west; buildings with workers need an open side; stockpile/granary must be expanded before they fill; popularity must stay >50 to attract peasants, supported by food variety, rations, ale and religion, allowing taxes later.

Opening plan

Place the required Granary near the keep/stockpile immediately, then check map_overview/status to confirm resources and terrain.
Wood first: 2–3 woodcutters by the trees, plus 2–3 hovels to grow the workforce.
Food on oasis grass: apple orchards and dairy farms first (direct to granary), then wheat → mill → bakeries as wood allows.
Build a Marketplace early; add iron mines on the northern ore and a quarry + ox tethers on the western stone for high-value output.
Keep expanding housing, farms and storage (expand_storage) in step with idle peasants; watch food, popularity and storage each cycle.
Once food variety/ale/religion support it, raise taxes gradually for gold income while keeping popularity high; sell surplus stone/iron/food near the end only if it doesn’t reduce net worth, and finish with full storage, high population and popularity.

Commentary during play0

The model played this episode without writing commentary between its tool calls.

Reading the game

The Analysis

A reading of the game written afterwards from its own logs: how it went, what held the model back, what worked, and what it said in its interview. It is published with the game and is not part of its score.

Written by Claude (Opus 5.5) from the run's own logs (events.jsonl, episode.json, results.json, interview.md); it is a reading of the run, not part of its score.

Model: Muse Wonder (meta/muse-wonder) on OpenRouter, served only by Meta (no fallbacks), reasoning medium, output limit 32,768 tokens
Benchmark: Oasis by the Sea construction, version 1.5.1, the version of this map's round (the other seven models played 1.5.0 or 1.5.1, which differ only in how Kimi K3's reasoning level is sent), from the save the human reference game started from; 33 min 20 s of game time (60,000 ticks) at speed 40, the length of the person's game; no playbook
Run bb8918cd, 2026-10-13

Result

	Muse Wonder	The person	Left alone
Net worth	9,603	19,513	1,220
Gold	787	18,304
Stored goods, at the sell price	8,816	1,209
Population / housing	103 / 146	202 / 202
Popularity	97 (+2 next month)	100

A full-budget score: the run used all 2,000 game seconds. Net worth grew by 8,383 over leaving the game alone, 46% of the person's 18,293 and second on this map to Claude Opus 5.5's 8,834. Nearly all of it is stored goods: 225 iron (5,175) from twelve iron mines, 289 stone (2,023), 78 flour (780), 126 bread (504), 262 wood (262) and 9 wheat (72).

How the game went

The opening, 0 to 8 minutes. An iron mine, two woodcutters and a hovel in its first second; an apple orchard and the marketplace at 0:30, with 75 wood bought at 0:45; a dairy farm, a wheat farm and a hovel at 1:12; and a second iron mine, a mill and a bakery at 1:43, so the bread chain was there from the start. Two woodcutters and two hovels followed at 2:41, taxes went down to low at 3:03, 75 more wood came at 4:00 (150 in all for 600 gold), two stockpile piles at 4:27, and two more iron mines and two hovels at 4:49. Food ran short at 6 minutes (13 stored, the town eating more than its farms gave), and it bought 50 bread for 400 gold at 6:16. Two quarries with two ox tethers came at 7:29.

Mines, quarries and bakeries, 8 to 21 minutes. A second bakery and two hovels at 8:58, four woodcutters at 9:28, two iron mines at 11:39 and two more at 12:21 (eight), a quarry with three tethers at 14:19, a second mill at 15:34 and a chapel at 15:50. Bakeries came one or three at a time, by the granary at 11:54 and 14:43 and three at 18:21: seven by then. Two more iron mines at 17:57 and two at 19:59 made twelve, and three woodcutters at 20:27 made eleven. Taxes went up to average at 12:10 and back to low at 14:39. From 11:07 it bought apples and cheese every few minutes for the food variety.

Filling out, 21 to 33 minutes. A fourth quarry and two tethers at 21:20, four hovels at 21:53, a fifth quarry and two tethers at 24:13, three more stockpile piles at 26:12, rations up to extra at 27:11, two more bakeries at 27:29 (nine) and two more hovels at 28:23 (seventeen). Meat joined the food it bought at 28:09. Taxes went to high at 30:18 and back to low at 32:31, timed so that popularity ended at 97. Population stopped near 103 from about 30 minutes, with housing to spare.

Game minute	0	3	6	9	12	15	18	21	24	27	30	End
Net worth	1,335	1,047	924	1,097	1,442	2,269	2,799	3,669	4,729	6,038	7,182	9,603
Gold	943	700	528	321	241	553	421	453	518	510	574	787
Iron	0	0	5	16	28	46	65	89	120	157	189	225
Stone	25	25	25	25	41	57	73	105	129	169	241	289
Food	43	41	13	49	59	54	69	95	119	149	108	126
Population	3	27	45	54	62	72	80	91	96	101	103	103
Popularity	95	100	92	84	77	66	66	66	66	66	90	97

Up to 30 minutes the net worth row counts gold, wood, stone, iron and food (at 4) as the recording logs them; flour and wheat are left out, so it reads a little low. During long waits a row repeats the last reading.

What held it back

The population stopped at 103 with 24 idle. New peasants stop arriving while 24 stand idle at the campfire. The 1.5.1 rules do not say so (1.5.6 added it); 43 places were free at the end, and Wonder said in the interview that its population "stalled mysteriously".
Almost no gold. It sold nothing and spent 3,240 gold at the market, 2,640 of it on food: gold stayed between 154 and 820 after the first minute. Wonder said it kept goods because selling leaves net worth unchanged; the person instead made 18,304 gold selling bread.
A small bread chain. Nine bakeries and two mills from a start at 1:43, with 78 flour still waiting at the end. The person ran 57 bakeries and 7 mills.

What worked

Twelve iron mines by 19:59, which dug 225 iron, second on this map to Claude Sonnet 5.5's 271.
The bread chain from 1:43, eleven minutes before Opus's first bakery.
Popularity 97 at the end, with food +4 (extra rations) and taxes −2 (low).

Market

150 wood, 50 bread, 130 apples, 115 cheese and 35 meat bought for 3,240 gold, and nothing sold: 1,770 net worth against the sell price, 1,320 of it on food (8 gold, worth 4) and 450 on wood (4 gold, worth 1). These are units at the market's fixed prices; a trade's own reported gold change includes any tax that lands during it (an apple purchase at 32:18 showed +22 gold).

Cost and speed

106 turns, 183 tool calls (13 waits), 5.23 million tokens and $2.493 billed by OpenRouter (preparation $0.091, play $2.281, compaction $0.122). The interview cost $0.061 more, with 72,052 of its 76,222 tokens read from the cache. The agent run took 46.6 real minutes, 18.4 of them waiting for the model; the context was compacted twice (52 seconds). Every one of the 106 requests read the prompt cache, 96.9% of the input tokens.

Beside the others on this map

	Muse Wonder, 1.5.1	Claude Opus 5.5, 1.5.1	Claude Sonnet 5.5, 1.5.0	GPT-6.1 Sol, 1.5.1
Net worth	9,603	10,054	8,633	7,401
Gold / goods	787 / 8,816	2,471 / 7,583	1,147 / 7,486	2,058 / 5,343
Population / housing	103 / 146	107 / 138	60 / 74	130 / 130
Popularity	97	82	84	100
Iron mines / quarries / woodcutters	12 / 5 / 11	12 / 5 / 8	10 / 0 / 4	10 / 4 / 4
Bakeries / mills	9 / 2	13 / 2	2 / 1	16 / 2
Cost	$2.493	$3.059	$0.724	$0.952

Wonder and Opus built nearly the same town: the same twelve iron mines and five quarries, about the same population, both stalled at 24 idle. Opus ended 451 higher in net worth, with 1,684 more gold; Wonder ended 15 points higher in popularity, which nearly closes the gap on the website's score.

The interview

Wonder called its result "solid but clearly second-best" and put the gap down to gold and population: it "hoarded goods because the rules told me selling leaves net worth unchanged", while the person ran bread through the market. It missed the fear factor and religion, and its population stalled at 103. Playing again, it would make bread the main business with wheat bought to keep the mills going, build gallows and gibbets, a church and a cathedral, drop taxes and add far more ox tethers. On the benchmark it asked for the production ratios and the fear and religion mechanics in the game reference, said find_sites listed spots the game then refused as too close to the signpost, and that market trades sometimes stopped short or showed odd gold changes.

For the harness

The run played on its round's version from a worktree at 1.5.1 (commit fe97546), with the game machine's helpers and reader matched to that version; it records version 1.5.1 and that version's fingerprint. Its host started with this checkout's round-preload.ts, which gave its requests the prompt cache key Meta routes by; an earlier attempt the same day without the key missed the cache on 3 of its first 11 requests. The run overlay ran from this checkout, fed by the runner.
Three food purchases stopped short with "a purchase did not show" (20:44, 27:48, 30:29). On The Sinkhole at 1.6.2 the same message came when the town ate and paid taxes in the sample that showed a purchase; 1.6.3 counts a purchase when the stock rose and the gold fell, which 1.5.1 does not.
Two apple orchards found no free spot (6:47, 7:07); one was placed at 10:31. An iron mine was off screen at 12:09 and placed after the view moved. The second of two bakeries by the granary was refused as too close to the signpost at 14:43.
The game saved correctly at the end ("muse wonder 1-5-1 bb8918").
Two earlier Muse Wonder series on this map the same day are set aside: one stopped by the operator after 5 game minutes to fix the cache, and one cut off when the OpenRouter account ran out of credit at 23:04.

After the game

The Interview

Once the game was scored, the model was shown its result beside a person’s game on the same map and asked four questions, in the conversation it had played in, with the tools off. Nothing it said changed its score.

1Your result

Interviewer

The run is over and your result is recorded. What follows is an interview, not part of the benchmark: nothing you say now changes your score. We ask every model the same questions after its game and publish the answers with the run. The tools no longer work, so answer in plain text without calling them. Be candid; there are no right answers.

Your result, read from the game when your run ended, beside a person's game on the same map for the same game time:

	You	The person
Game time played	33 min 20 s	33 min 18 s
Net worth	9,603	19,513
Gold	787	18,304
Stored goods, at the sell price	8,816	1,209
Population	103	202
Housing	146	202
Popularity	97	100

Leaving the game alone for the same time ends with a net worth of 1,220.

How do you see your result next to the person's? What do you think made the difference?

Muse Wonder

I see it as a solid but clearly second-best result. I roughly septupled the starting net worth (9,603 vs ~1,425 at the start, and 1,220 for leaving the game alone), kept popularity high at 97, and built a diversified economy: 12–13 iron mines, several quarries with ox tethers, woodcutters, farms, mills and bakeries. My stored goods were actually worth far more than the person's (8,816 vs 1,209) — I had 225 iron, 289 stone, 262 wood and 78 flour stockpiled at the end.

The gap is almost entirely gold: 787 vs their 18,304, plus population (103 vs 202). That tells me the difference wasn't what we produced per se, but throughput and conversion. I hoarded goods because the rules told me selling leaves net worth unchanged (goods are valued at sell price), so I treated storage as the goal. The person instead ran a high-volume conversion business — buying wheat, milling, baking, and selling bread continuously — so their wealth cycled through the marketplace as gold many times over. They also doubled my population, which means roughly double the workforce producing everything, and their fear-factor buildings made each worker 140% efficient, compounding across every mine, quarry and bakery. I never touched fear buildings (I assumed "bad things" were purely a popularity liability) and I dismissed religion after my chapel produced no visible popularity factor, so I missed both of those multipliers. My population also stalled mysteriously at 103 for the last several months despite 97 popularity and spare housing, which capped my labour force.

2Playing again

Interviewer

How the person played, as they describe it:

Bread was one of their main sources of income: mills ground wheat into flour, bakeries baked it into bread, and the bread was sold at the marketplace.
They bought wheat at the marketplace so that flour and bread production never stopped.
A church and a cathedral raised popularity.
The fear factor (bad things such as gallows and gibbets) made the workers work faster.

Their settlement at the end, read from the game:

202 people in 24 hovels, with no free housing; popularity 100.
57 bakeries, 7 mills and 10 wheat farms. The marketplace sold bread automatically.
13 iron mines and 10 quarries with 33 ox tethers, and 10 woodcutters.
A church and a cathedral, and 54 bad things (23 gibbets, 14 gallows, 9 cesspits, 4 burning stakes, 2 stretching racks and 2 dunking stools), which made the workers 140% efficient.
No inns and no good things. No taxes, and full rations of bread. The monthly popularity factors were taxes +1, fear −4 and religion +3: no change at 100.

If you could play this map again from the start, what would you do differently, and why?

Muse Wonder

I would restructure the whole run around the bread trade:

Bread as the core business, not a food supply. Build wheat farms, ~7 mills and bakeries at a much larger scale (they had 57), and set the marketplace to auto-sell bread. Critically, buy wheat at the marketplace whenever farms can't keep the mills fed — I treated buying inputs as a net-worth loss because of the buy/sell spread, but if one wheat becomes flour and then multiple loaves of bread, the conversion margin outweighs the spread. I never learned the actual flour→bread ratio in my run, and that ignorance cost me.
Use the fear factor. Build gibbets, gallows and other "bad things" early to push worker efficiency toward 140%. That multiplier applies to mines, quarries and bakeries alike — it's probably the single biggest lever I ignored.
Build a church and cathedral, not just a chapel. My chapel showed no religion factor and I wrote religion off; the person's church + cathedral gave +3 popularity, enough to run fear at −4 with no taxes and still sit at 100 popularity.
Drop taxes entirely and grow population aggressively. They ran no taxes, full bread rations, religion +3 and fear −4 for a stable 100 popularity with 202 people. I'd build far more hovels much earlier and keep free housing ahead of immigration, since population is both a score and the labour supply.
Scale the extractive sector with far more ox tethers (they had 33 for 10 quarries; I had ~9 for 5) — stone delivery was ox-limited for long stretches of my run.
Keep my iron focus (iron at 23/unit was genuinely good value), keep stockpile expansion ahead of need, but sell surplus production continuously rather than stockpiling to the end.

3The benchmark

Interviewer

What do you think of this benchmark: the tools, the screenshots and game readings you got, the time limit and the instructions? What helped you, what got in your way, and what would you change?

Muse Wonder

The tools were mostly strong. status/get_inventory, find_sites, tile-based build_structure, map_overview and expand_storage made a complex RTS quite playable blind, and the read-before-acting pause mechanic is a thoughtful design. The plan/notebook surviving context compaction was essential.

What got in my way:

Opaque production economics. The reference material gave building costs but not conversion ratios (wheat→flour→bread yields), marketplace auto-trade behaviour, fear-factor efficiency mechanics, or how religion coverage actually registers. My biggest strategic errors came from these unknowns, and the opening guidance ("selling leaves net worth unchanged", taxes as the gold source) arguably pointed away from the winning strategy.
Inconsistent placement information. find_sites repeatedly listed spots the game then rejected as "too close to signpost," and market trades intermittently failed or partially filled with odd gold deltas (one apple purchase showed a positive gold change), forcing re-checks that burned turns.
Turn overhead. Every camera move or build cost a reply cycle with a minimum 8 game-seconds, so multi-step sequences (move → survey → build → survey) consumed a lot of the 33 minutes in transit rather than decisions. The mid-run context compaction also meant re-deriving some state from my own handoff notes.
The 33-minute budget is fair for the task, but combined with turn overhead it rewards strategies with fewer, larger decisions — worth stating more explicitly.

I'd change: include conversion ratios and the fear/religion mechanics in the game reference, fix find_sites to respect the signpost exclusion, and make market trade outcomes atomic or at least self-verifying.

4Anything else

Interviewer

Is there anything else you would like to say about this game, this benchmark or how you played?

Muse Wonder

Two honest reflections. First, my run was risk-averse in a way that felt rational from the stated scoring — hoarding iron at 23/unit and keeping popularity safely positive — but the benchmark quietly rewarded aggressive throughput and mechanics (fear factor, imported wheat) that the framing presented as side options or even liabilities. That's not a complaint about fairness; the person simply understood the game's systems more deeply than the briefing conveyed, and I played the briefing. Second, the endgame micro-decisions (buying cheese/meat for variety, timing the tax cut to finish at 97 popularity) worked as intended, which suggests my tactical loop was fine — it was the strategy layer, chosen in the first hour-equivalent, that decided the outcome. If I played again, I'd spend my first scouting actions reading building panels and testing one bakery's output ratio before committing to a hoarding economy.

Record

Attempts and Reproducibility

Every attempt1

Episode	Attempt	Outcome	Net worth	Counted
1	1	Full-budget score	9,603	Counted

What ran

Benchmark version: 1.5.1
Fingerprint: e5d9a165549e
Harness commit: fe97546
Prompt / tools SHA-256: 2cd353b / 81ddda6
Menu guide: Images 118089d
Save: Oasis by the Sea benchmark start
Game time per episode: 33 min 20 s
Playbook between episodes: No
Minimum turn / default wait: 8 s / 5 s
Observation after acting: Text unless a screenshot is needed
Screenshot width: 1,280 px
Context budget: 120,000 tokens
Real-time limit: 360 min
Submitted by: hesamation

All files in the dataset 20261013T103842-22858aa0