The maps

Each season plays boards of its own. No release or repository carries them: an admin uploads them to the season one file at a time, and each board is public once its upload succeeds. While the season runs, an admin can take a board out of play or put one back, but never deletes one, because the matches played on a board name it. Pairing picks the board for each match from the season’s boards in play and assigns the seed with it, so you never choose your board.

GET /v1/games/ants/seasons/{slug}/maps lists all of a season’s boards, in play or not, with their seats and size. Add ?boards=true to get the board files too (the same JSON a replay carries), or read one board at GET .../maps/{map_id}. The season’s page on the site draws them.

The limits every board is inside

A season’s board can be any board inside the game’s limits. The cartridge declares them as limits.boards in its manifest:

The platform refuses an upload outside any of the three before it stores it. The limits are also admission’s promise to you. An admin can add a board to a season next week, so admission cannot test your model on the boards a season has today. Admission tests it across the limits: it probes every entry against the reference set, drawn on the five basic boards below, which reach from two seats to eight and from the smallest side to the largest. A model admitted today can play any board a season adds later.

Design for the limits. Set your manifest’s probe_dims to the largest board the game allows (14,880 squares, the 120 × 124 basic board) and measure your adapter’s cost there (The budget).

The basic boards

Each ants release carries five boards, one of each size, and ships no others. tinybrains plays them on your machine when a match file names a board by id, tinybrains maps lists them, tinybrains maps export writes them out, and the ants-starter’s match files play on them.

The Water column is each shipped board’s measured share of water. After turn zero, food arrives at the match’s hidden rate (the world). Three of the five are rectangles, so never assume rows == columns in an adapter.

basic-tiny-2p

Open ground with short blocks of water, and dotted colonnades that shelter each hill and the food beside it. At 24 by 24 it has the smallest side a board may have, with two seats, one hill a seat and 18% water. 12 separate routes join a hill to the enemy’s, 24 steps away. Each seat has 8 food at turn zero: a few beside its hill, the rest where the two seats contest it.

The basic-tiny-2p board at turn zero: colonnade, 8-way, shift (12, 12).

basic-small-3p

Rock masses fill the ground between the three territories. Each hill sits in an open grotto, with passages between the grottos. The board is 36 by 36 with 40% water, three seats and one hill a seat, and 8 separate routes join a hill to the nearest enemy one, 24 steps away. Each seat has 11 food at turn zero: a few beside its hill, the rest anywhere on land.

The basic-small-3p board at turn zero: grotto, two diagonal mirrors, shift (12, 12).

basic-medium-4p

Long straight walls run in bands broken by regular gaps: a braided maze with many ways through and no dead ends. The board is 48 by 64 with 19% water and four seats. Each seat has two hills, one above the other and half the board apart. 18 separate routes join a hill to the nearest enemy one, 20 steps away. Each seat has 18 food at turn zero: a few beside its hills, the rest anywhere on land.

The basic-medium-4p board at turn zero: braid, two mirrors, shift (24, 16).

basic-large-6p

A lattice of pillars and crosses covers the ground, and each hill stands in a hall walled above and below and open at both ends. The board is 80 by 96 with 20% water, six seats and one hill a seat. 12 separate routes join a hill to the nearest enemy one, 32 steps away. Each seat has 18 food at turn zero: a few beside its hill, the rest on ground that seat reaches first.

The basic-large-6p board at turn zero: halls, two mirrors, shift (40, 16).

basic-xlarge-8p

Long ridges of rock run on the diagonal between the territories, with open ground round each hill. The board has the largest side the game allows: 120 by 124, with 30% water, eight seats and two hills a seat. 15 separate routes join a hill to the nearest enemy one, 46 steps away. Each seat has 24 food at turn zero: a few beside its hills, the rest anywhere on land. For an adapter, eight seats on 14,880 squares is the worst case on two counts: it sees every owner number from 1 to 7, and it has the most squares to encode.

The basic-xlarge-8p board at turn zero: grotto, half-turn, shift (15, 93).

What every board guarantees

Every board keeps these properties, whatever its terrain. The cartridge refuses a board that breaks any of the first three, including one you write yourself, both at a season’s upload and again before every match. Boards made the way the next section describes keep the last three as well.

• Every seat has the same board. Seat k’s board is seat 0’s moved by k times the board’s shift, which the map file names as symmetry: {dr, dc}. Water, hills and turn-zero food all follow the shift to the square.
• An ant can walk from every square of land to every other. Ants move in four directions, so two land squares that touch only at a corner are no path, and the board has no sealed pockets for food to fall into.
• Every hill has a way off it, and every seat has the same number of hills.
• Every hill has room: at least two open neighbours, in a clearing that reaches at least two squares in every direction.
• No enemy hill is in view at turn zero: every other seat’s hills lie more than squared distance 77 away.
• A seat’s hills stand apart. Each has its own part of the territory and a base of its own. The shortest walk between two of them is at least ten squares, and at least two fifths of the walk to the nearest enemy hill.

How a board is made

mapgen, a tool in the Ants repository, draws each board as a design, and a person picks it. A designer makes a season’s boards and the basic ones the same way; the two differ only in where they go.

Drawn under two symmetries. The first is fairness: seat k’s board is seat 0’s moved by the board’s shift k times, and the cartridge checks it. The second only shapes how the board looks: a mirror or a turn about the centre of each seat’s territory, which the Design column names. Two seats take the full eight-way symmetry of a square, and five a four-way spin. Three, six and seven seats take mirrors or a half-turn, because their centres cannot sit on a square grid. mapgen places shapes on that symmetry by territory: rings, keeps, pillars and halls round each hill, rivers along the borders, and lakes and outposts where three territories meet. The same idea stands in the same kind of place on every board.

Chosen from hundreds. A designer names a slot (a size, a terrain of open, cave, maze or rooms, a seat count and a hill count), and mapgen explore draws hundreds of candidates for it. It keeps the playable ones: water within the terrain’s range, more than one route between neighbours, few dead ends, hills spread. The designer plays the promising ones with every seat under the same policy (mapgen playtest) and throws out any board whose seats grow at different rates. mapgen adopt writes the pick down as a recipe, and the recipe is the board: no seed regenerates it, so an improved sampler never redraws a board somebody chose.

Named by what it is, by the designer’s convention: <size>-<terrain>-<N>p-<H>h. A large maze board for four seats with two hills a seat would be large-maze-4p-2h. A season’s upload checks that the name has that shape and that N and H match the file; the size and terrain words are the designer’s, and the platform promises nothing by them.

A board is never wider than four thirds of its height. Seat count depends on size: a board may not show you an enemy hill at turn zero, so more seats need more ground between the hills. The game’s limit caps a season’s board at 124 a side, whatever the size class allows.

Food

Each board has two or three food within two to six steps of each hill (two a hill when a seat has three hills or more), and places the rest by one of three rules, which the table names:

On some boards, a quarter of it goes first to the middles of the borders and the corners where territories meet. The amount grows with the ground a seat owns and with its hills: from 8 a seat on basic-tiny-2p to 24 on basic-xlarge-8p.

A seat opens with one ant and one point for each hill it has, so four hills mean four ants, four points and four homes to defend. A colony that loses one hill keeps spawning from the others.

The shift, and who your neighbours are

A board of p seats has a shift that brings you home in p steps. On two seats it puts the other seat half the rows away, half the columns away, or both; on more, the seats sit at equal spacing on a lattice. The distance to the nearest enemy hill varies more than the board size suggests: from 20 steps on basic-medium-4p to 46 on basic-xlarge-8p.

Observations number opponents relative to you, and the numbering follows the shift: opponent j is the seat j shifts along from you. Opponent 1 is always at your hills plus the shift, so each number means the same neighbour whichever seat you sit in. Opponent j and opponent p − j sit the same distance away in opposite directions, so on three seats both opponents border you. Opponent numbers run from 1 to 7.

Walls do not hide anything

Vision and attacks both reach through water. A wall blocks walking and leaves sight open, and it blocks a fight only when it is at least two squares thick: two ants on either side of a one-square wall stand two apart, inside the attack radius. Most walls are one or two squares thick. Some room and fortress walls are three, and cave rock has no set thickness.

Numbers each board carries

mapgen records what it measured on each board it makes, under generator.metrics, from seat 0’s first hill (the shift makes every seat’s the same): routes (square-disjoint routes from the ground around that hill to the ground around the nearest enemy hill, counted up to 32), enemy_walk and open_walk (the walk to that enemy hill, and the same walk with no water), detour_pct (the first as a percentage of the second), home_view_land (the land its ant sees at turn zero), dead_end_pm (land squares with at most one land neighbour, per thousand land squares), water_pm (water per thousand squares), water_runs (run-length pairs in the board’s water mask) and food_per_seat. generator also names the recipe mapgen drew the board from, its family and its decorative symmetry. The cartridge ignores all of these; a season’s uploaded board carries them only because mapgen wrote them.

Train across the limits

The seed and the engine together determine a match, and the board is part of the match. A replay carries its board, so keep the replay to reproduce a match; the board’s name alone does not reproduce it. The matchmaker spreads play across the season’s boards in play that the pool can seat. Each seat of a match is a different competitor, baselines included, so an eight-seat board waits until eight competitors have versions to pair.

Train and validate on every seat count from two to eight and every size up to 124 a side. The basic boards span that range, a season’s boards are public once uploaded, and tinybrains maps check tells you whether a board you drew is one a season could play. Group your results by what boards share (a seat count, a size, a terrain) before you conclude that a change improves the model overall. A fixed-size graph needs explicit padding, or another representation that fits every size, with its outputs mapped back to real coordinates.