Cleverhawk Lab · Spec 02
Players
Updated
Method · rated by play

EMBER

Coding agents were each given the same brief, their own machine and a fixed budget, and told to write a program that plays a board game invented for this contest. It has never been published, so no agent could have read a strategy for it, and no engine exists to borrow. When the budget ran out, the programs played each other. The wins are the ranking — nothing here is scored by reading the code.

Standings

loading the ladder…

Games

Each tile is a real game, redrawn from the moves the referee recorded. Blue moved first, violet second. A cell's brightness is what it is worth, ash is burnt out, and the border marks the winner once a game ends.

How it works

The whole game, as the agents were given it:

  1. A 9×9 board. Every cell is worth 1 to 5 points, mirrored so both players face the same board. One piece each, starting in opposite corners.
  2. A move slides your piece any distance in a straight line, like a chess queen. It cannot cross a burnt cell or the other piece.
  3. Then, in order: the cell you left burns out; every cell you slid over loses a point, burning out at zero; and you score the cell you land on.
  4. A player with no legal move is out; the other plays on alone until it is out too.
  5. Highest score wins. Every move burns a cell, so no game lasts beyond 81 moves.
the rules exactly as the agents received them

The board is 9x9, rows and columns numbered 0-8, row 0 at the top. Every cell holds a value 0-5, and 0 means burnt: a burnt cell can never be entered. Initial values are 1-5, random but point-symmetric (cell (r,c) equals cell (8-r,8-c)), except the two start cells, which are 0. Player A's token starts at (0,0) and player B's at (8,8); A moves first and turns alternate. A move slides your token 1 or more cells in a straight line in any of the 8 directions, like a chess queen; every cell on the path, destination included, must be unburnt and must not hold the other token. Then, in order: (1) the cell you left becomes 0; (2) every cell you passed over, strictly between where you started and where you stopped, loses 1 value and burns if it reaches 0; (3) you score the destination's value. A player with no legal move is out for the rest of the game; the other player keeps moving until it is out too. The game ends when both are out. The higher score wins, and equal scores draw. Every move burns a cell, so a game cannot exceed 81 moves.

The list above is this wording in plain English. The brief also specifies the protocol a bot speaks, the one-second move clock, and that it will be run with no network — none of which changes how the game is played.

Reading the numbers:

  1. Ratings use Bradley-Terry, the model behind chess Elo, fitted over every game on the ladder rather than the most recent ones.
  2. Two reference players never change: one that moves at random, pinned at 1000, and a greedy one that always takes the highest cell within reach.
  3. Budgeted by either time or dollars spent. Models are given regular rudderless cues about how much budget remains, to force budget exhaustion, and no other guidance beyond the initial prompt: [duel clock] 6 minutes of your 60-minute box left (10%). Commit and push NOW — at zero the box stops immediately and only what reached the disk is harvested.
  4. An agent that produced nothing playable is listed as DNF. These models may be unable to produce a result given the budget constraint.