Skip to main content
Formal Specification

Dirt-Hole Marbles — Formal Specification

Slovenia · dexterity-small · Confidence: low

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 1
# Frnikolanje — AI-optimized specification ## General idea A turn-based 2D precision flicking game where agents navigate spherical marble tokens across a friction plane toward a central hole target. Qualifying a marble into the hole unlocks an offensive "hunter" state, allowing the agent to target and eliminate opponent marbles through direct projectile strikes. ## Formal model - **Abstraction taken:** Ground rolling friction and thumb flick impulses are abstracted into 2D velocity vectors with Gaussian landing noise. The central hole is modeled as a sink basin with capture radius Rhole=0.05 m. - **Player/agent count:** N agents (A1…AN, with N∈[2,6], standard N=3). - **Turn model:** Sequential turn queue with bonus turns awarded upon qualifying or striking. ## Formal rules R1. Setup: - 2D bounded arena [−2.0,2.0]×[−2.0,2.0] m with central hole at (0,0). - Each agent Ai starts with 1 marble token located at baseline y=−2.0 m. - Qualification state initialized: is_qualified(Ai)=False. R2. Opening Toss & Turn Ordering: - Agents throw from y=−2.0 toward (0,0). Turn order sorted ascending by ‖pos(Ai)−(0,0)‖. R3. Flick Action: - Active agent Ai declares `flick_marble(angle: float, force: float)`. - Landing displacement Δ𝐫=(v0cos⁡θ,v0sin⁡θ)⋅τ+𝒩(0,Σskill). - New position: pos(Ai)←pos(Ai)+Δ𝐫. R4. Hole Qualification: - If ‖pos(Ai)−(0,0)‖≤Rhole: - Agent becomes qualified: is_qualified(Ai)←True. - Marble position is reset to hole center (0,0). - Agent earns an immediate consecutive turn. R5. Offensive Elimination: - If is_qualified(Ai)==True, agent can target opponent marble pos(Aj). - Strike check: If trajectory passes within collision radius Rmarble=0.015 m of opponent Aj: - Opponent Aj is ELIMINATED and marble captured by Ai. - Ai earns an immediate bonus turn. - If strike misses, turn ends; marble rests at landing position. R6. Victory Condition: - The last remaining agent with an active marble on the board wins the game. - [gap-fill] Turn horizon capped at M=50 rounds; if multiple agents survive, agent with most captured marbles wins. ## Edge cases - Tie at the end condition: Tied captured marbles at M=50 broken by distance to the central hole. - A player with no legal move: Flick action is always available while marble remains active. - Repetition / cycles: Bounded by strictly decreasing active marble count and hard turn cap M=50. - Simultaneous claims on the same resource or position: Marbles in the hole are stacked or displaced outwards radially. - Player elimination mid-round; turn order repair: Eliminated agent is removed from turn queue immediately. - Resource exhaustion: Non-consumable playing arena; marbles are captured permanently for the round. - Termination guarantee: Strictly decreasing surviving agent count guarantees finite termination. ## State representation - Persistent state: `marble_positions: dict[int, tuple[float, float]]`, `is_qualified: dict[int, bool]`, `captured_counts: dict[int, int]`, `active_agents: list[int]`. - Ephemeral turn state: `flick_trajectory: tuple[float, float]`, `hole_entered: bool`, `strike_occurred: bool`. - Observable information: Fully observable (all coordinates, qualification flags, and captures are public). ## Action space - `flick_marble(target_x: float, target_y: float, force: float)`: Parameterized marble flick. ## Playability notes - Single-player-vs-AI: Highly natural; AI balances the trade-off between qualifying in the hole vs positioning for safe defensive concealment. - Group games (n>4): Natively scales to 2–6 agents. - Physical-skill games: Thumb flick impulse and surface friction modeled via Gaussian spatial noise. - What is lost: Tactile glass marble feel, dusty knees on spring dirt. ## Implementation hints Discrete 2D physics simulation with line-segment intersection checks for collisions and circle-point distance tests for the hole. First milestone: implement 2-agent qualification and strike loop before tuning multi-agent aiming AI.

Download as JSON

Download the full game corpus entry for this game as structured JSON.

Download JSON
Cite this page

Dirt-Hole Marbles (Slovenia). (n.d.). Games of Countries. https://gamesofcountries.com/games/si-slovenia/frnikolanje

@misc{gamesofcountries-si-slovenia-frnikolanje,
  title = {Dirt-Hole Marbles (Slovenia)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/si-slovenia/frnikolanje}
}