Snake & Ladder 100K Tournaments
A tournament-structured simulation of 16 AI agents, each with a distinct decision-making style drawn from human psychology, over 3.1 million games.
What it contains
What this is
A behavioural simulation of 16 AI agents playing Snake & Ladder, each with a decision style drawn from human psychology, run through 100,000 complete tournaments: about 3.1 million games.
Tournament structure
Each tournament mimics a real competition, so results carry genuine selection pressure rather than flat pairwise noise:
- Group stage: four groups of four, round-robin, top two advance
- Quarter-finals: 8 agents → 4
- Semi-finals: 4 → 2
- Final: 2 → 1 champion
The 16 agents
| Agent | Profile | Behaviour |
|---|---|---|
| maximizer | Greedy achiever | Always picks the higher of two rolls |
| minimizer | Conservative risk-avoider | Always picks the lower roll |
| smart_maximizer | Calculated strategist | Seeks ladders, avoids snakes where possible |
| skipper | Snake-fearful | Skips a turn if a snake is ahead |
| bounded_skipper | Rational risk manager | Skips up to three turns to avoid danger |
| ladder_seeker | Opportunist | Prioritises landing on ladders |
| snake_avoider | Loss-averse | Avoids every snake, even when suboptimal |
| random_choice | Fate-trusting | Picks between two rolls at random |
| always_first | Impulsive | Always takes the first roll |
| always_second | Chronic second-guesser | Always takes the second roll |
| steady_climber | Stability-seeker | Avoids big jumps, prefers steady gains |
| impatient | Big-move seeker | Skips small rolls, waits for high ones |
| gambler | Risk-taking recalibrator | Rerolls a mediocre first result |
| adaptive | Flexible realist | Aggressive early, cautious near the end |
| ladder_waiter | Delay-gratifier | Skips unless the exact roll hits a ladder |
| bounded_ladder_waiter | Patient but practical | Waits for ladders, with a skip limit |
The data
A single match-level CSV (tournament_results.csv): one row per game, with tournament number, round, group, both players, winner and number of turns.
Use it for
Agent decision-making under chance, tournament dynamics, strategy dominance, and probabilistic modelling. A companion release, 80M games with compact move encoding, logs every move for 10 of the strategies.
DOI: 10.34740/kaggle/dsv/12648398 · Licence: CC BY 4.0 · Code: GitHub
Agents
16
Tournaments
100,000
Games
3.1M
Kaggle usability
10.0