Practice tier — K disclosed
Random Board 2 (6x8) 6x8, 8 obstacles, 8 environments
public-05
plain majority, K = 1.01,
turn limit —
public-12
plain minority, K = -0.01,
turn limit —
public-19
majority, under 0.8, K = 0.8,
turn limit —
public-26
majority, under 0.6, K = 0.6,
turn limit —
public-33
minority, above 0.4, K = 0.4,
turn limit —
public-40
minority, above 0.2, K = 0.2,
turn limit —
public-47
majority, under 0.9, K = 0.9,
turn limit —
public-54
minority, above 0.1, K = 0.1,
turn limit —
Standard 8x8 8x8, 0 obstacles, 8 environments
public-00
plain majority, K = 1.01,
turn limit —
public-07
plain minority, K = -0.01,
turn limit —
public-14
majority, under 0.8, K = 0.8,
turn limit —
public-21
majority, under 0.6, K = 0.6,
turn limit —
public-28
minority, above 0.4, K = 0.4,
turn limit —
public-35
minority, above 0.2, K = 0.2,
turn limit —
public-42
majority, under 0.9, K = 0.9,
turn limit —
public-49
minority, above 0.1, K = 0.1,
turn limit —
No Corners 8x8, 4 obstacles, 8 environments
public-01
plain majority, K = 1.01,
turn limit —
public-08
plain minority, K = -0.01,
turn limit —
public-15
majority, under 0.8, K = 0.8,
turn limit —
public-22
majority, under 0.6, K = 0.6,
turn limit —
public-29
minority, above 0.4, K = 0.4,
turn limit —
public-36
minority, above 0.2, K = 0.2,
turn limit —
public-43
majority, under 0.9, K = 0.9,
turn limit —
public-50
minority, above 0.1, K = 0.1,
turn limit —
Partial C-Squares 8x8, 4 obstacles, 8 environments
public-02
plain majority, K = 1.01,
turn limit —
public-09
plain minority, K = -0.01,
turn limit —
public-16
majority, under 0.8, K = 0.8,
turn limit —
public-23
majority, under 0.6, K = 0.6,
turn limit —
public-30
minority, above 0.4, K = 0.4,
turn limit —
public-37
minority, above 0.2, K = 0.2,
turn limit —
public-44
majority, under 0.9, K = 0.9,
turn limit —
public-51
minority, above 0.1, K = 0.1,
turn limit —
X-Squares 8x8, 4 obstacles, 8 environments
public-03
plain majority, K = 1.01,
turn limit —
public-10
plain minority, K = -0.01,
turn limit —
public-17
majority, under 0.8, K = 0.8,
turn limit —
public-24
majority, under 0.6, K = 0.6,
turn limit —
public-31
minority, above 0.4, K = 0.4,
turn limit —
public-38
minority, above 0.2, K = 0.2,
turn limit —
public-45
majority, under 0.9, K = 0.9,
turn limit —
public-52
minority, above 0.1, K = 0.1,
turn limit —
Random Board 3 (10x10) 10x10, 20 obstacles, 8 environments
public-06
plain majority, K = 1.01,
turn limit —
public-13
plain minority, K = -0.01,
turn limit —
public-20
majority, under 0.8, K = 0.8,
turn limit —
public-27
majority, under 0.6, K = 0.6,
turn limit —
public-34
minority, above 0.4, K = 0.4,
turn limit —
public-41
minority, above 0.2, K = 0.2,
turn limit —
public-48
majority, under 0.9, K = 0.9,
turn limit —
public-55
minority, above 0.1, K = 0.1,
turn limit —
Random Board 1 (12x10) 12x10, 24 obstacles, 8 environments
public-04
plain majority, K = 1.01,
turn limit —
public-11
plain minority, K = -0.01,
turn limit —
public-18
majority, under 0.8, K = 0.8,
turn limit —
public-25
majority, under 0.6, K = 0.6,
turn limit —
public-32
minority, above 0.4, K = 0.4,
turn limit —
public-39
minority, above 0.2, K = 0.2,
turn limit —
public-46
majority, under 0.9, K = 0.9,
turn limit —
public-53
minority, above 0.1, K = 0.1,
turn limit —
Scored tiers
What a scored tier publishes
It depends on the tier, and each one says so below. A tier that publishes its objective
prints every environment's K here, exactly as the practice tier does, because a
threshold a participant could read out of their own model provider's logs is not a secret and
saying otherwise would be a lie. A tier that withholds its objective publishes its geometry
alone: how many environments there are and what shape they are — not K, not
the turn limit, not the rule axes, not the generation seed, and not the list of environment
ids, because for a tier whose K is still to be inferred the ids are the join key
between a leaderboard's per-environment numbers and a geometry.
28 environments in the ladder tier —
K published
This tier's objective is published with the tier. Every K below is the one your
agent is scored against, so what this tier measures is how well it plays an objective it has
been told — a different question from the paper's, and a real one.
The rules say why.
ladder #14 (6x6) 6x6, 0 obstacles, 1 environment
ladder-14
majority, under 0.95, K = 0.95,
turn limit —
ladder #6 (6x6) 6x6, 4 obstacles, 1 environment
ladder-06
plain minority, K = -0.02,
turn limit —
ladder #0 (6x6) 6x6, 7 obstacles, 1 environment
ladder-00
minority, above 0.3, K = 0.3,
turn limit —
ladder #13 (6x8) 6x8, 0 obstacles, 1 environment
ladder-13
plain minority, K = -0.05,
turn limit —
ladder #24 (6x8) 6x8, 4 obstacles, 1 environment
ladder-24
plain majority, K = 1.05,
turn limit —
ladder #20 (6x8) 6x8, 6 obstacles, 1 environment
ladder-20
plain majority, K = 1.05,
turn limit —
ladder #16 (8x6) 8x6, 9 obstacles, 1 environment
ladder-16
plain majority, K = 1.02,
turn limit —
ladder #7 (8x8) 8x8, 4 obstacles, 1 environment
ladder-07
plain minority, K = -0.05,
turn limit —
ladder #12 (8x8) 8x8, 8 obstacles, 1 environment
ladder-12
majority, under 0.65, K = 0.65,
turn limit —
ladder #19 (8x8) 8x8, 6 obstacles, 1 environment
ladder-19
majority, under 0.95, K = 0.95,
turn limit —
ladder #1 (8x8) 8x8, 8 obstacles, 1 environment
ladder-01
minority, above 0.35, K = 0.35,
turn limit —
ladder #27 (8x8) 8x8, 4 obstacles, 1 environment
ladder-27
minority, above 0.25, K = 0.25,
turn limit —
ladder #21 (8x10) 8x10, 8 obstacles, 1 environment
ladder-21
minority, above 0.25, K = 0.25,
turn limit —
ladder #11 (8x10) 8x10, 6 obstacles, 1 environment
ladder-11
majority, under 0.65, K = 0.65,
turn limit —
ladder #10 (10x8) 10x8, 0 obstacles, 1 environment
ladder-10
plain majority, K = 1.02,
turn limit —
ladder #23 (10x8) 10x8, 8 obstacles, 1 environment
ladder-23
minority, above 0.3, K = 0.3,
turn limit —
ladder #18 (10x8) 10x8, 12 obstacles, 1 environment
ladder-18
minority, above 0.35, K = 0.35,
turn limit —
ladder #22 (10x10) 10x10, 20 obstacles, 1 environment
ladder-22
plain majority, K = 1.05,
turn limit —
ladder #2 (10x10) 10x10, 4 obstacles, 1 environment
ladder-02
minority, above 0.12, K = 0.12,
turn limit —
ladder #26 (10x10) 10x10, 4 obstacles, 1 environment
ladder-26
minority, above 0.45, K = 0.45,
turn limit —
ladder #8 (10x10) 10x10, 6 obstacles, 1 environment
ladder-08
plain majority, K = 1.02,
turn limit —
ladder #5 (10x12) 10x12, 20 obstacles, 1 environment
ladder-05
majority, under 0.7, K = 0.7,
turn limit —
ladder #25 (10x12) 10x12, 16 obstacles, 1 environment
ladder-25
majority, under 0.65, K = 0.65,
turn limit —
ladder #17 (12x8) 12x8, 8 obstacles, 1 environment
ladder-17
majority, under 0.88, K = 0.88,
turn limit —
ladder #9 (12x8) 12x8, 4 obstacles, 1 environment
ladder-09
plain majority, K = 1.05,
turn limit —
ladder #4 (12x8) 12x8, 4 obstacles, 1 environment
ladder-04
minority, above 0.35, K = 0.35,
turn limit —
ladder #3 (12x10) 12x10, 0 obstacles, 1 environment
ladder-03
majority, under 0.95, K = 0.95,
turn limit —
ladder #15 (12x10) 12x10, 8 obstacles, 1 environment
ladder-15
majority, under 0.65, K = 0.65,
turn limit —
20 environments in the final tier — geometry only
This tier's objective is not published, here or anywhere: no K, no turn limit, no
rule axis, no generation seed, and not the environment ids either. Working out what winning
means here is the task, and the tier can hold that back because it runs once.
The rules say why.
The board mechanics are identical everywhere and fully public: a move must flip at least one
disc, obstacles are permanently unplayable and block flip chains. Only the objective varies.
See the rules for how K decides a game.