04 / THE CONDITIONS
Methodology
Rules and opponents
The models played under the published Catan base rules using the open-source catanatron engine. First to 10 victory points wins. The discard limit is 7. Domestic trade uses a small menu of affordable 1:1 and 2:1 swaps, with repeat offers limited.
Qualifier: two games against three random-move bots, in seats 1 and 4. A win in either game is a pass; all models still enter the league. League: six rounds, three tables per round. Top eight advance to two semifinal tables, then a four-player final with at most three games. First to two wins is champion; three different winners can leave no champion. Boss ladder: VictoryPoint, ValueFunction, then AlphaBeta; stop at the first loss, with one tiebreak for a one-point loss.
Shared state budget
State, instruction and options share an estimated 6,000-token limit, within 75% of the smallest 8,192-token context. Only the oldest history is removed. The board, own hand, public opponent information and scores remain. Opponent resource types and hidden victory cards stay private to the players. Replays reveal hidden state to viewers.
Choosing and rotating
Argmax means picking the option with the highest returned probability. Ties use option order. Oversized menus use separate kind and target questions; their odds are conditional, not a distribution over the full original menu.
Seats rotate. Six league games cannot divide equally across four seats: each model uses two seats twice and two seats once. Opponent pairs are balanced as closely as the schedule allows. Each round shares a board seed.
Failures and the turn guard
Transport failures allow up to three attempts; malformed answers allow two. Rate limits and server failures use backoff. Permanent client errors stop retrying. A failed decision may use a random legal fallback, which is flagged. Five consecutive failures pause a game. Forced moves skip inference. At 250 turns, actual points decide the winner, with seat order breaking a tie. Paused games do not count as losses.
Engine: catanatron, GPL-3.0, commit ecf9311. Each replay retains the full commit and version.
The instruction, verbatim
You are playing Catan to win (first to 10 victory points). Choose the best move.
Framing v1: short options, verbatim example
Each menu depends on the current board. This is the first recorded model decision for this framing. All other inputs are in the game logs.
Instruction
You are playing Catan to win (first to 10 victory points). Choose the best move.
State
You are Red. Turn 0. Phase: BUILD_INITIAL_SETTLEMENT.
Road costs 1 Wood + 1 Brick. Settlement costs 1 Wood + 1 Brick + 1 Sheep + 1 Wheat; it must connect to your road and stay two edges from every settlement/city. City costs 2 Wheat + 3 Ore. Development card costs 1 Sheep + 1 Wheat + 1 Ore. Settlement earns 1 point, city 2, longest road (at least 5) and largest army (at least 3 Knights) earn 2 each. A roll produces 1 card per settlement or 2 per city on matching tiles, except the robber tile. Pips count dice combinations out of 36. On 7, hands over 7 discard half, then move robber and steal. You may build or trade repeatedly before ending your turn. New development cards cannot be played this turn. Knight and robber movement are separate decisions. Opponents' resource types and hidden victory cards are unknown.
Board:
Tile 0: Brick 11, 2 pips; cube (0, 0, 0); corners 0,1,2,3,4,5
Tile 1: Ore 10, 3 pips; cube (1, -1, 0); corners 6,7,8,9,2,1
Tile 2: Sheep 3, 2 pips; cube (0, -1, 1); corners 2,9,10,11,12,3
Tile 3: Wheat 6, 5 pips; cube (-1, 0, 1); corners 4,3,12,13,14,15
Tile 4: Wood 5, 4 pips; cube (-1, 1, 0); corners 16,5,4,15,17,18
Tile 5: Ore 4, 3 pips; cube (0, 1, -1); corners 19,20,0,5,16,21
Tile 6: Brick 9, 4 pips; cube (1, 0, -1); corners 22,23,6,1,0,20
Tile 7: Wood 5, 4 pips; cube (2, -2, 0); corners 24,25,26,27,8,7
Tile 8: Sheep 8, 5 pips; cube (1, -2, 1); corners 8,27,28,29,10,9
Tile 9: Brick 4, 3 pips; cube (0, -2, 2); corners 10,29,30,31,32,11
Tile 10: Wheat 11, 2 pips; cube (-1, -1, 2); corners 12,11,32,33,34,13
Tile 11: Sheep 12, 1 pips; cube (-2, 0, 2); corners 14,13,34,35,36,37
Tile 12: Wheat 9, 4 pips; cube (-2, 1, 1); corners 17,15,14,37,38,39
Tile 13: Wood 10, 3 pips; cube (-2, 2, 0); corners 40,18,17,39,41,42
Tile 14: Sheep 8, 5 pips; cube (-1, 2, -1); corners 43,21,16,18,40,44
Tile 15: Ore 3, 2 pips; cube (0, 2, -2); corners 45,46,19,21,43,47
Tile 16: Desert, 0 pips; cube (1, 1, -2); corners 48,49,22,20,19,46
Tile 17: Wood 6, 5 pips; cube (2, 0, -2); corners 50,51,52,23,22,49
Tile 18: Wheat 2, 1 pips; cube (2, -1, -1); corners 52,53,24,7,6,23
Robber: tile 16 (Desert).
Pieces: none
Roads: none
Ports: Sheep 2:1 at corners [25, 26]; Any resource 3:1 at corners [28, 29, 32, 33, 35, 36, 38, 39]; Ore 2:1 at corners [40, 44]; Wheat 2:1 at corners [45, 47]; Brick 2:1 at corners [48, 49]; Wood 2:1 at corners [52, 53]
Red: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Your hand: Wood 0, Brick 0, Sheep 0, Wheat 0, Ore 0
Your development cards: Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Your total points including hidden victory cards: 0
Blue: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
White: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Orange: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Bank: Wood 19, Brick 19, Sheep 19, Wheat 19, Ore 19; development cards remaining 25
Last 20 events:
Options
a0: Settlement at corner 0: Brick 11, Ore 4, Brick 9
a1: Settlement at corner 10: Sheep 3, Sheep 8, Brick 4
a2: Settlement at corner 11: Sheep 3, Brick 4, Wheat 11
a3: Settlement at corner 12: Sheep 3, Wheat 6, Wheat 11
a4: Settlement at corner 13: Wheat 6, Wheat 11, Sheep 12
a5: Settlement at corner 14: Wheat 6, Sheep 12, Wheat 9
a6: Settlement at corner 15: Wheat 6, Wood 5, Wheat 9
a7: Settlement at corner 16: Wood 5, Ore 4, Sheep 8
a8: Settlement at corner 17: Wood 5, Wheat 9, Wood 10
a9: Settlement at corner 18: Wood 5, Wood 10, Sheep 8
a10: Settlement at corner 19: Ore 4, Ore 3, Desert
a11: Settlement at corner 1: Brick 11, Ore 10, Brick 9
a12: Settlement at corner 20: Ore 4, Brick 9, Desert
a13: Settlement at corner 21: Ore 4, Sheep 8, Ore 3
a14: Settlement at corner 22: Brick 9, Desert, Wood 6
a15: Settlement at corner 23: Brick 9, Wood 6, Wheat 2
a16: Settlement at corner 24: Wood 5, Wheat 2
a17: Settlement at corner 25: Wood 5
a18: Settlement at corner 26: Wood 5
a19: Settlement at corner 27: Wood 5, Sheep 8
a20: Settlement at corner 28: Sheep 8
a21: Settlement at corner 29: Sheep 8, Brick 4
a22: Settlement at corner 2: Brick 11, Ore 10, Sheep 3
a23: Settlement at corner 30: Brick 4
a24: Settlement at corner 31: Brick 4
a25: Settlement at corner 32: Brick 4, Wheat 11
a26: Settlement at corner 33: Wheat 11
a27: Settlement at corner 34: Wheat 11, Sheep 12
a28: Settlement at corner 35: Sheep 12
a29: Settlement at corner 36: Sheep 12
a30: Settlement at corner 37: Sheep 12, Wheat 9
a31: Settlement at corner 38: Wheat 9
a32: Settlement at corner 39: Wheat 9, Wood 10
a33: Settlement at corner 3: Brick 11, Sheep 3, Wheat 6
a34: Settlement at corner 40: Wood 10, Sheep 8
a35: Settlement at corner 41: Wood 10
a36: Settlement at corner 42: Wood 10
a37: Settlement at corner 43: Sheep 8, Ore 3
a38: Settlement at corner 44: Sheep 8
a39: Settlement at corner 45: Ore 3
a40: Settlement at corner 46: Ore 3, Desert
a41: Settlement at corner 47: Ore 3
a42: Settlement at corner 48: Desert
a43: Settlement at corner 49: Desert, Wood 6
a44: Settlement at corner 4: Brick 11, Wheat 6, Wood 5
a45: Settlement at corner 50: Wood 6
a46: Settlement at corner 51: Wood 6
a47: Settlement at corner 52: Wood 6, Wheat 2
a48: Settlement at corner 53: Wheat 2
a49: Settlement at corner 5: Brick 11, Wood 5, Ore 4
a50: Settlement at corner 6: Ore 10, Brick 9, Wheat 2
a51: Settlement at corner 7: Ore 10, Wood 5, Wheat 2
a52: Settlement at corner 8: Ore 10, Wood 5, Sheep 8
a53: Settlement at corner 9: Ore 10, Sheep 3, Sheep 8
Framing v2: options with facts, verbatim example
Each menu depends on the current board. This is the first recorded model decision for this framing. All other inputs are in the game logs.
Instruction
You are playing Catan to win (first to 10 victory points). Choose the best move. Points come from settlements, cities, longest road, largest army and victory cards; expanding to new corners grows production.
State
You are Red. Turn 0. Phase: BUILD_INITIAL_SETTLEMENT.
Road costs 1 Wood + 1 Brick. Settlement costs 1 Wood + 1 Brick + 1 Sheep + 1 Wheat; it must connect to your road and stay two edges from every settlement/city. City costs 2 Wheat + 3 Ore. Development card costs 1 Sheep + 1 Wheat + 1 Ore. Settlement earns 1 point, city 2, longest road (at least 5) and largest army (at least 3 Knights) earn 2 each. A roll produces 1 card per settlement or 2 per city on matching tiles, except the robber tile. Pips count dice combinations out of 36. On 7, hands over 7 discard half, then move robber and steal. You may build or trade repeatedly before ending your turn. New development cards cannot be played this turn. Knight and robber movement are separate decisions. Opponents' resource types and hidden victory cards are unknown.
Board:
Tile 0: Brick 11, 2 pips; cube (0, 0, 0); corners 0,1,2,3,4,5
Tile 1: Ore 10, 3 pips; cube (1, -1, 0); corners 6,7,8,9,2,1
Tile 2: Sheep 3, 2 pips; cube (0, -1, 1); corners 2,9,10,11,12,3
Tile 3: Wheat 6, 5 pips; cube (-1, 0, 1); corners 4,3,12,13,14,15
Tile 4: Wood 5, 4 pips; cube (-1, 1, 0); corners 16,5,4,15,17,18
Tile 5: Ore 4, 3 pips; cube (0, 1, -1); corners 19,20,0,5,16,21
Tile 6: Brick 9, 4 pips; cube (1, 0, -1); corners 22,23,6,1,0,20
Tile 7: Wood 5, 4 pips; cube (2, -2, 0); corners 24,25,26,27,8,7
Tile 8: Sheep 8, 5 pips; cube (1, -2, 1); corners 8,27,28,29,10,9
Tile 9: Brick 4, 3 pips; cube (0, -2, 2); corners 10,29,30,31,32,11
Tile 10: Wheat 11, 2 pips; cube (-1, -1, 2); corners 12,11,32,33,34,13
Tile 11: Sheep 12, 1 pips; cube (-2, 0, 2); corners 14,13,34,35,36,37
Tile 12: Wheat 9, 4 pips; cube (-2, 1, 1); corners 17,15,14,37,38,39
Tile 13: Wood 10, 3 pips; cube (-2, 2, 0); corners 40,18,17,39,41,42
Tile 14: Sheep 8, 5 pips; cube (-1, 2, -1); corners 43,21,16,18,40,44
Tile 15: Ore 3, 2 pips; cube (0, 2, -2); corners 45,46,19,21,43,47
Tile 16: Desert, 0 pips; cube (1, 1, -2); corners 48,49,22,20,19,46
Tile 17: Wood 6, 5 pips; cube (2, 0, -2); corners 50,51,52,23,22,49
Tile 18: Wheat 2, 1 pips; cube (2, -1, -1); corners 52,53,24,7,6,23
Robber: tile 16 (Desert).
Pieces: none
Roads: none
Ports: Sheep 2:1 at corners [25, 26]; Any resource 3:1 at corners [28, 29, 32, 33, 35, 36, 38, 39]; Ore 2:1 at corners [40, 44]; Wheat 2:1 at corners [45, 47]; Brick 2:1 at corners [48, 49]; Wood 2:1 at corners [52, 53]
Red: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Your hand: Wood 0, Brick 0, Sheep 0, Wheat 0, Ore 0
Your development cards: Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Your total points including hidden victory cards: 0
Blue: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
White: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Orange: 0 public points; 0 resource cards; 0 development cards; road length 0; holds longest road False; holds largest army False; played Knight 0, Year_Of_Plenty 0, Monopoly 0, Road_Building 0, Victory_Point 0
Bank: Wood 19, Brick 19, Sheep 19, Wheat 19, Ore 19; development cards remaining 25
Last 20 events:
Options
a0: Settlement at corner 0: adds Brick 11 (2 pips), Ore 4 (3 pips), Brick 9 (4 pips); your points after 1
a1: Settlement at corner 10: adds Sheep 3 (2 pips), Sheep 8 (5 pips), Brick 4 (3 pips); your points after 1
a2: Settlement at corner 11: adds Sheep 3 (2 pips), Brick 4 (3 pips), Wheat 11 (2 pips); your points after 1
a3: Settlement at corner 12: adds Sheep 3 (2 pips), Wheat 6 (5 pips), Wheat 11 (2 pips); your points after 1
a4: Settlement at corner 13: adds Wheat 6 (5 pips), Wheat 11 (2 pips), Sheep 12 (1 pips); your points after 1
a5: Settlement at corner 14: adds Wheat 6 (5 pips), Sheep 12 (1 pips), Wheat 9 (4 pips); your points after 1
a6: Settlement at corner 15: adds Wheat 6 (5 pips), Wood 5 (4 pips), Wheat 9 (4 pips); your points after 1
a7: Settlement at corner 16: adds Wood 5 (4 pips), Ore 4 (3 pips), Sheep 8 (5 pips); your points after 1
a8: Settlement at corner 17: adds Wood 5 (4 pips), Wheat 9 (4 pips), Wood 10 (3 pips); your points after 1
a9: Settlement at corner 18: adds Wood 5 (4 pips), Wood 10 (3 pips), Sheep 8 (5 pips); your points after 1
a10: Settlement at corner 19: adds Ore 4 (3 pips), Ore 3 (2 pips); your points after 1
a11: Settlement at corner 1: adds Brick 11 (2 pips), Ore 10 (3 pips), Brick 9 (4 pips); your points after 1
a12: Settlement at corner 20: adds Ore 4 (3 pips), Brick 9 (4 pips); your points after 1
a13: Settlement at corner 21: adds Ore 4 (3 pips), Sheep 8 (5 pips), Ore 3 (2 pips); your points after 1
a14: Settlement at corner 22: adds Brick 9 (4 pips), Wood 6 (5 pips); your points after 1
a15: Settlement at corner 23: adds Brick 9 (4 pips), Wood 6 (5 pips), Wheat 2 (1 pips); your points after 1
a16: Settlement at corner 24: adds Wood 5 (4 pips), Wheat 2 (1 pips); your points after 1
a17: Settlement at corner 25: adds Wood 5 (4 pips); your points after 1
a18: Settlement at corner 26: adds Wood 5 (4 pips); your points after 1
a19: Settlement at corner 27: adds Wood 5 (4 pips), Sheep 8 (5 pips); your points after 1
a20: Settlement at corner 28: adds Sheep 8 (5 pips); your points after 1
a21: Settlement at corner 29: adds Sheep 8 (5 pips), Brick 4 (3 pips); your points after 1
a22: Settlement at corner 2: adds Brick 11 (2 pips), Ore 10 (3 pips), Sheep 3 (2 pips); your points after 1
a23: Settlement at corner 30: adds Brick 4 (3 pips); your points after 1
a24: Settlement at corner 31: adds Brick 4 (3 pips); your points after 1
a25: Settlement at corner 32: adds Brick 4 (3 pips), Wheat 11 (2 pips); your points after 1
a26: Settlement at corner 33: adds Wheat 11 (2 pips); your points after 1
a27: Settlement at corner 34: adds Wheat 11 (2 pips), Sheep 12 (1 pips); your points after 1
a28: Settlement at corner 35: adds Sheep 12 (1 pips); your points after 1
a29: Settlement at corner 36: adds Sheep 12 (1 pips); your points after 1
a30: Settlement at corner 37: adds Sheep 12 (1 pips), Wheat 9 (4 pips); your points after 1
a31: Settlement at corner 38: adds Wheat 9 (4 pips); your points after 1
a32: Settlement at corner 39: adds Wheat 9 (4 pips), Wood 10 (3 pips); your points after 1
a33: Settlement at corner 3: adds Brick 11 (2 pips), Sheep 3 (2 pips), Wheat 6 (5 pips); your points after 1
a34: Settlement at corner 40: adds Wood 10 (3 pips), Sheep 8 (5 pips); your points after 1
a35: Settlement at corner 41: adds Wood 10 (3 pips); your points after 1
a36: Settlement at corner 42: adds Wood 10 (3 pips); your points after 1
a37: Settlement at corner 43: adds Sheep 8 (5 pips), Ore 3 (2 pips); your points after 1
a38: Settlement at corner 44: adds Sheep 8 (5 pips); your points after 1
a39: Settlement at corner 45: adds Ore 3 (2 pips); your points after 1
a40: Settlement at corner 46: adds Ore 3 (2 pips); your points after 1
a41: Settlement at corner 47: adds Ore 3 (2 pips); your points after 1
a42: Settlement at corner 48: adds none; your points after 1
a43: Settlement at corner 49: adds Wood 6 (5 pips); your points after 1
a44: Settlement at corner 4: adds Brick 11 (2 pips), Wheat 6 (5 pips), Wood 5 (4 pips); your points after 1
a45: Settlement at corner 50: adds Wood 6 (5 pips); your points after 1
a46: Settlement at corner 51: adds Wood 6 (5 pips); your points after 1
a47: Settlement at corner 52: adds Wood 6 (5 pips), Wheat 2 (1 pips); your points after 1
a48: Settlement at corner 53: adds Wheat 2 (1 pips); your points after 1
a49: Settlement at corner 5: adds Brick 11 (2 pips), Wood 5 (4 pips), Ore 4 (3 pips); your points after 1
a50: Settlement at corner 6: adds Ore 10 (3 pips), Brick 9 (4 pips), Wheat 2 (1 pips); your points after 1
a51: Settlement at corner 7: adds Ore 10 (3 pips), Wood 5 (4 pips), Wheat 2 (1 pips); your points after 1
a52: Settlement at corner 8: adds Ore 10 (3 pips), Wood 5 (4 pips), Sheep 8 (5 pips); your points after 1
a53: Settlement at corner 9: adds Ore 10 (3 pips), Sheep 3 (2 pips), Sheep 8 (5 pips); your points after 1
A/B pairs share initial seeds. Choices consume randomness, so later dice can diverge. This small experiment describes these games; it does not establish a general ranking. Confidence measures exclude bots, forced moves and split questions. Cost per game uses completed seat-games; total costs also include incomplete games. Latency is per model decision, including retries.