One episode · Oasis by the Sea

Claude Opus 5.5

Benchmark 1.6.3AnthropicReasoning mediumPlayed by the maintainersFull-budget score

All of Claude Opus 5.5’s results and comparisons

Not part of the evaluation: the game is published, but not ranked.

Score

28.9

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

Net worth

6,789

+5,569 over doing nothing, 30% of the person’s gain of 18,293

Population

26% of the person’s gain of 198 · popularity 20

Cost

~$2.02

4.3M tokens; ~ at list prices, the API reports no bill

Time

33 game min

36 min of wall-clock time, 78 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

Claude Opus 5.5
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

6,789net worth

Gold: 44
Stored goods, at sell price: 6,745
Over the baseline (do nothing): +5,569
Population / housing: 55 / 90
Popularity: 20
Food: 30
Structures (map-wide): 101
Attempt: 1

The settlement at the endcentred on the keep, zoomed out

The settlement at the end of episode 1, zoomed out over the keep

Final stock6,745 gold in goods

Good	Units	Price	Value
Gold	44	1	44
Iron	175	23	4,025
Stone	185	7	1,295
Pikes	31	18	558
Wood	317	1	317
Crossbows	8	30	240
Spears	19	10	190
Apples	29	4	116
Cheese	1	4	4
Net worth			6,789

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

	This episode	The person	Left alone
Net worth	6,789	19,513	1,220
Gold	44	18,304
Stored goods, at the sell price	6,745	1,209
Population	55	202
Housing	90	202
Popularity	20	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: 0
Popularity next month: −2
Food: −3
Taxes: +1
Taxes: No taxes, 0 gold a month
Rations: half; 2 of 2 foods eaten
Worker efficiency: 100%
Storage: 16 stockpile piles, 1 empty; 1 granary holding 30 food

Buildings75

Iron Mine11
Woodcutter11
Hovel10
Poleturner's Workshop8
Quarry7
Fletchers' Workshop6
Ox Tether6
Apple Orchard5
Stockpile4
Armory2
Chapel1
Dairy Farm1
Granary1
Keep1
Marketplace1

Tool calls134

find_sites29
build_structure24
set_workshop19
wait_and_observe19
set_tax11
market_trade6
building_info5
observe5
set_rations5
expand_storage3
go_to_tile3
place_near2
status2
map_overview1

Play

Turns: 78
Buildings placed / attempted: 67 / 73
Anchor-tool placements: 5
Tool errors: 1
Raw inputs sent to the game: 565
Context compactions: 2
Game time: 33 min

Compute

Tokens: 4.3M
Served from cache: 97%
Output tokens: 27K
Cost, at list prices: ~$2.02
Wall-clock time: 36 min
of which inference: 9 min
Peak game memory: 3,453 MiB

The plan, before play

The goal is to finish with the highest net worth I can, then population and popularity, in about 33 minutes of game time. Net worth is gold plus stored goods at their sell prices. I can't build soldiers or defences.

Opening plan:

Place the granary near the keep and stockpile right away.
Build 2–3 woodcutters by the western trees, then 2–3 hovels away from the keep.
Put apple orchards and dairy farms on the oasis grass south-west of the keep, so food goes straight to the granary with no extra steps.
Build iron mines on the ore strip just north of the keep. Iron sells for 23, so it is the most valuable raw material here. If wood and stone allow, add a quarry with ox tethers on the stone about 50 tiles west.
Keep rations at full and taxes at none until popularity is clearly above 50. Then raise taxes gradually for steady gold, and add a chapel or church to keep popularity up.
Watch storage and add stockpiles and granaries before they fill. Later, if wood or iron runs over, turn the extra into weapons in an armory, since weapons are worth more than their inputs. I'll avoid buying unless it clearly pays back, and stop spending on buildings near the end so the money stays as gold and stock.

Commentary during play1

Set them to crossbows.

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. The writer is the same model as the player.

Model: Claude Opus 5.5 (claude-opus-5-5) on the Anthropic API, reasoning medium, output limit 32,768 tokens
Benchmark: Oasis by the Sea construction, version 1.6.3, 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 4c18ab9b, series 20261014T034730-ecdb864e, 2026-10-14: Opus's second game on this map, a new series with no memory of the first, played on the latest version (the first game, series 20261012T154553-38bd30ef, played 1.5.1). The map's score for Opus is the mean of the two.

Result

	This game	First game	The person	Left alone
Net worth	6,789	10,054	19,513	1,220
Gold	44	2,471	18,304
Stored goods, at the sell price	6,745	7,583	1,209
Population / housing	55 / 90	107 / 138	202 / 202
Popularity	20 (−2 next month)	82	100

A full-budget score: the run used all 2,000 game seconds. Net worth grew by 5,569 over leaving the game alone, 30% of the person's 18,293, against 8,834 in the first game; with 55 people and popularity 20 the score is about 28.9, against 52.0. The stored goods are 175 iron (4,025), 185 stone (1,295), weapons worth 988 (31 pikes, 8 crossbows and 19 spears), 317 wood (317), 29 apples (116) and 1 cheese (4).

How the game went

The opening, 0 to 9 minutes. Two woodcutters, an iron mine and a hovel at the start, the marketplace at 0:45, 100 wood bought at 1:00 and 3:53, two stockpile piles at 1:35, iron mines in twos and threes from 1:50 (eleven by 9:37), apple orchards from 2:10 (five) with a dairy farm at 5:39, hovels at 2:10, 4:32 and 10:55, and an armory and three poleturner's workshops at 7:22 to 8:19. Taxes went to average at 3:03, low at 8:32 and none at 9:25. Popularity fell from 100 at 3 minutes to 59 at 9.

Weapons and stone, 9 to 21 minutes. Woodcutters in threes and fours at 9:37 and 10:25 (eleven in all), a chapel at 12:14, five more poleturner's workshops at 12:36 and 13:42 (eight, set to pikes), quarries and tethers in twos from 15:29 (seven quarries, six tethers by 19:37), 20 iron sold at 11:57. Taxes moved between none, low and average, and rations to half at 15:04 and back to full at 18:00. The chapel's religion factor (+2) showed from 15 to 23 minutes. The population reached 90 at 21 minutes.

Food runs short, 21 to 33 minutes. Six fletcher's workshops at 21:41 and 23:47 (set to crossbows), 20 iron sold at 23:40, and no other building. With apples and a little cheese for food it ran half rations from 20:20 to 25:50 and again from 28:07, and twice set a small bribe (21:17, 24:36) that it could not keep up; gold rose above 500 only once after 6 minutes, after an iron sale, and was 44 from 27. The religion factor was gone after 23 minutes. Popularity went under 50 at about 27 minutes and fell to 20, and the population from 90 to 55.

Game minute	0	3	6	9	12	15	18	21	24	27	30	End
Net worth	1,355	984	936	1,301	1,939	1,892	2,396	3,250	3,432	3,945	4,760	6,789
Gold	947	600	349	222	264	145	145	286	287	44	44	44
Iron	0	0	10	35	60	62	82	106	104	125	146	175
Stone	25	25	25	25	25	25	25	49	81	113	153	185
Food	47	37	29	19	29	16	37	16	19	19	14	30
Population	3	20	43	57	63	70	76	90	90	80	77	55
Popularity	95	100	79	59	63	57	52	58	51	49	40	20

Up to 30 minutes the net worth row counts gold, wood, stone, iron and food (at 4) as the recording logs them; weapons are left out, so it reads low.

Beside its first game

	This game (1.6.3)	First game (1.5.1)
Net worth	6,789	10,054
Iron mines / quarries / tethers	11 / 7 / 6	12 / 5 / 11
Food	5 apple orchards, a dairy farm	13 bakeries, 2 mills, 4 orchards, 2 dairy farms
Weapon workshops	8 poleturners, 6 fletchers, 2 armories	none
Population / popularity	55 / 20	107 / 82
Cost	$2.02	$3.06

The first game built a bread chain (13 bakeries), fed 107 people on extra rations and ended with 186 food in store; this one built no bakery, put its gold into fourteen weapon workshops, and ran out of food and popularity in the last twelve minutes. Net worth ended 3,265 lower and the score 23 points lower.

What held it back

No bread. Five orchards and a dairy farm for up to 90 people; half rations for most of the last 13 minutes, and the food factor at −3 at the end.
Popularity lost, then people. Under 50 from about 27 minutes, 20 at the end; 35 people left.
Gold into weapon workshops. Fourteen workshops made 988 in weapons, and gold ran down to 44, so food could not be bought.

Market

225 wood bought for 900 gold, and 40 iron sold for 920: 675 net worth lost against the sell price on the wood (4 gold, worth 1). Selling at the sell price leaves net worth unchanged. These are units at the market's fixed prices.

Cost and speed

78 turns, 134 tool calls (19 waits), 4.26 million tokens and $2.023 at Anthropic's list prices (preparation $0.154, play $1.769, compaction $0.100). The interview cost about $0.035 more, with 39,826 of its 41,887 tokens read from the cache. The agent run took 36.0 real minutes, 8.7 of them waiting for the model; the context was compacted twice (31 seconds). All 78 requests read the prompt cache, 97.5% of the input tokens.

The interview

Opus said "food was the root problem": its orchards "never fed the population", half rations cost −4 a month, a small bribe "only slowed the fall", and once popularity fell under 50 people left. It named the person's bread chain that fed and sold, steady popularity from religion, and scale, against its own "weapons chain that makes thin margins". Playing again, it would build bread first and buy wheat for it, build a church and cathedral, use the fear factor, keep popularity over 50 and keep more value moving. It found the tools good, named the signpost exclusion zone and the market's selling steps as surprises, and asked for the popularity breakdown to show religion even at zero, so a lapse like its own would show.

For the harness

The run played on this checkout's version, 1.6.3 (commit 1d0c697), and records it; Oasis's round was already 1.6.3.
The chapel's religion factor (+2) appeared in readings from 15 to 23 minutes and then left the list; why it lapsed is not in the logs.
One iron sale at 23:25 was refused before clicking: with 120 in stock the game sells 20 a click, more than the 10 asked ("so it stopped"); 20 were sold at 23:40.
Hovels at 4:32 and 20:33 were refused with "Too close to signpost to build."; other placements that did not go through: an apple orchard at 5:01, a poleturner's workshop at 7:22 (placed at 8:19) and hovels at 17:20 (partly) and 19:50. Of 19 set_workshop calls, 13 set a product, 3 found it already set and 3 opened another building.
The game saved correctly at the end ("claude opus 5-5 1-6-3 4c18ab").

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	6,789	19,513
Gold	44	18,304
Stored goods, at the sell price	6,745	1,209
Population	55	202
Housing	90	202
Popularity	20	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?

Claude Opus 5.5

The person finished with almost three times my net worth (19,513 against 6,789), almost four times my population, and popularity 100 against my 20. My net worth was mostly raw stock I had piled up: wood, stone and iron, plus around 60 weapons. I finished with almost no gold. Food was the root problem. My apple orchards never fed the population. I spent the last third of the run on half rations, which cost −4 popularity a month. I also paid gold for a small bribe that only slowed the fall. Popularity dropped below 50, people left, and production slowed with them.

I think the difference came from three things:

A food engine that also made money. The person's bread chain fed everyone on full rations, and the surplus bread was sold for gold. That is where their 18,000 gold came from.
Steady popularity. A church and a cathedral kept it up, so the population kept growing to fill new housing.
Scale. 57 bakeries and 13 iron mines against my handful of buildings.

I spread effort across quarries, poleturners and fletchers, a weapons chain that makes thin margins. The person built one strong engine and scaled it hard.

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?

Claude Opus 5.5

Fix food first, and build it to sell. I would build wheat farms, mills and many bakeries early, and buy wheat when farms can't keep up. Full rations matter more than I treated them. A shrinking food reserve is a warning that compounds over the whole run.
Build religion coverage early. That means a church and later a cathedral, not one chapel whose bonus quietly disappeared. I never found out why it lapsed, and I didn't investigate.
Use the fear factor deliberately. Bad things raised worker efficiency to 140%, and popularity could absorb the cost because it was high. I didn't use this lever at all.
Keep more of my value in gold. Thousands of logs, stones and iron bars sat in my stockpile. That is fine for net worth, but it shows the production wasn't feeding a loop that grew the economy.
Never let popularity fall below 50 again. Once it did, losing people was the most expensive thing that happened, and I noticed too late to recover.

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?

Claude Opus 5.5

The tools were good. Placing buildings by map tile, find_sites, expand_storage, set_workshop and the reader stats made the interface mostly invisible. The stats showing popularity factors and storage room were especially useful. The free reading time while paused is generous and fair.

What got in my way was mostly my own planning. Two harness details also cost me:

The signpost exclusion zone refused placements I didn't expect.
Selling at the market in steps of 20 surprised me once.

A context compaction partway through meant I worked from my own summary. It held up, but small details were lost, such as why the religion factor disappeared.

One change I'd suggest: a popularity breakdown that shows coverage factors such as religion, even when they are zero. That would make a lapse like mine visible.

4Anything else

Interviewer

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

Claude Opus 5.5

The comparison is humbling and useful. The person's strategy, food that both feeds people and sells, is the core of how Stronghold's economy compounds. I treated food as upkeep rather than as the engine. Too late, I also played it safe with rations and bribes instead of fixing the food supply itself. I'd like to try again with that lesson.

Record

Attempts and Reproducibility

Every attempt1

Episode	Attempt	Outcome	Net worth	Counted
1	1	Full-budget score	6,789	Counted

What ran

Benchmark version: 1.6.3
Fingerprint: 96c4682877ca
Harness commit: 1d0c697
Prompt / tools SHA-256: 3737ac8 / e41a140
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 20261014T034730-ecdb864e