Formal Specification
Token Flipping — Formal Specification
Singapore · dexterity-small · Confidence: high
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 3
# Kuti-kuti — AI-optimized specification
## General idea
Kuti-kuti is a turn-based 2D spatial positioning and flicking duel on a bounded tabletop. Players alternate launching their plastic token toward the opponent's token, balancing aggressive attempts to land atop the enemy piece with defensive positioning to avoid leaving an easy counter-shot.
## Formal model
- **Abstraction taken:** The continuous finger flick is modeled as a 2D impulse vector with directional noise and distance decay on a bounded plane (default: cm).
- **Player/agent count:** Exactly 2 agents.
- **Turn model:** Strictly alternating sequential turns.
## Formal rules
R1. Setup: The table is a 2D bounding box . Player 1's token starts at and Player 2's token starts at . Each token is a circular disc of radius cm. Each player holds a reserve of tokens (default: ).
R2. Active Turn: On an agent's turn, the agent inputs a flick vector applied to their active token.
R3. Physics & Overlap Resolution:
- New position , where represents execution noise.
- If falls outside the boundary , the token is placed on the boundary edge at the exit point, and the active agent receives a 1-turn skip penalty [gap-fill].
- Capture check: If the Euclidean distance and the active token's trajectory landed from above onto the target disc, a capture occurs.
R4. Capture Consequence: The captured token is removed from play and awarded to the active player's score. If the defending player has remaining reserve tokens, a new token is deployed at their starting side. If no reserve remains, the round ends.
R5. Win Condition: The first agent to capture all of the opponent's tokens (or reach captures) wins the match. [gap-fill] A maximum turn cap of 50 turns per round prevents indefinite avoidance; if reached, the player with the closer position to the center wins.
## Edge cases
- Tie at the end condition: In the event of a turn-cap expiration with equal captures, the token closest to the geometric center wins.
- A player with no legal move: Not applicable; an active agent on the table always has the full action space of directional flicks.
- Repetition / cycles: Bounded by the 50-turn cap per round [gap-fill].
- Simultaneous claims on the same resource or position: Not applicable; turns alternate strictly, and overlap detection is evaluated only for the moving token upon coming to rest.
- Player elimination mid-round: When a token is captured, the player deploys their next reserve token; when reserves are 0, that player is eliminated and the match concludes.
- Resource exhaustion: Tokens are finite ( pieces per player).
- Termination guarantee: Guaranteed by the finite token count combined with the 50-turn cap per token match.
## State representation
- Persistent state:
- `scores`: array of size 2 (captured tokens count).
- `reserves`: array of size 2 (remaining tokens in hand).
- `turn_count`: integer .
- `active_player`: integer .
- `penalty_skip`: array of booleans `[p1_skip, p2_skip]`.
- Ephemeral state:
- `positions`: 2 2D coordinates .
- Observable information: Fully observable to both agents.
## Action space
- `flick(force, angle)`:
- `force` (scales max travel distance from 0 to 40 cm).
- `angle` radians.
## Playability notes
- Single-player-vs-AI: Highly natural; AI can use simple geometric heuristics (vector toward target with distance calibration) or minimax rollouts.
- Group games (n>4): Can be extended into a 4-player free-for-all or 2v2 tag team by adding token identifiers and rotating through active turns.
- Physical-skill games: Finger flicking accuracy is modeled as Gaussian noise scaled inversely by agent skill.
- What is lost: Three-dimensional aerodynamics of irregular plastic animal contours and tactical flipping techniques (e.g. edge-hopping).
## Implementation hints
Simple 2D continuous coordinates with simple circle-circle collision / overlap geometry. Fast execution suitable for Monte Carlo Tree Search (MCTS) or reinforcement learning.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Token Flipping (Singapore). (n.d.). Games of Countries. https://gamesofcountries.com/games/sg-singapore/kuti-kuti
@misc{gamesofcountries-sg-singapore-kuti-kuti,
title = {Token Flipping (Singapore)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/sg-singapore/kuti-kuti}
}