Formal Specification
Marbles (Mora variant) — Formal Specification
Israel · dexterity-small · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 4
# Gullot — AI-optimized specification
## General idea
Gullot (Marbles) is a competitive spatial dexterity game where agents flick physical spheres on a 2D surface. The goal is to first enter a central circular pit (the Mora) to unlock aggressive "Hunter" status, and then strike and eliminate opponents' spheres from the field. Strategy revolves around positional safety versus aggressive targeting.
## Formal model
- **Abstraction taken:** The 2D dirt field is modeled as a continuous Cartesian coordinate space with meters. The Mora is a circular target at origin with radius meters. Physical flicks are modeled as stochastic vectors where the agent commits an intended speed and angle, and the actual vector is sampled from a Gaussian distribution.
- **Player/agent count:** 2 to 6 players.
- **Turn model:** Sequential turn rotation. On their turn, an active agent can execute exactly one flick.
## Formal rules
R1. **Setup:** Each agent starts with their marble at a random point along the line (representing the throwing line).
R2. **State Variables:** Each agent has a persistent state `is_hunter` (boolean, defaults to false) and `is_active` (boolean, defaults to true).
R3. **The Flick Action:** On their turn, the active agent selects a target coordinate . The outcome position of their marble is calculated stochastically:
[gap-fill] The deviation from the intended path is proportional to the distance of the flick and inversely proportional to the agent's `accuracy_skill` parameter.
R4. **Entering the Mora:** If the final position falls within of , the marble is placed at , the player earns 1 point, and `is_hunter` is set to `true`.
R5. **Hunter Attacks:**
- If an active agent has `is_hunter == true`, they can target another agent's marble position .
- If their flicked marble intersects (within a collision radius meters), the target agent's `is_active` is set to `false` (eliminated).
- [gap-fill] After a successful strike, the Hunter's `is_hunter` is set to `false`, and they must re-enter the Mora to re-enable Hunter status.
R6. **Win Condition:** The game terminates when only one agent has `is_active == true`. That agent is declared the winner and receives a score equal to the count of eliminated opponents.
## Edge cases
- **Tie at the end condition:** Not applicable; collisions are resolved sequentially, so simultaneous elimination is impossible.
- **A player with no legal move:** Not applicable; any coordinate on the field is a valid flick target.
- **Repetition / cycles:** If players repeatedly miss the Mora or each other, the game could cycle. [gap-fill] To guarantee termination, after 30 total rounds of play, the Mora radius expands by 10% per round, ensuring players eventually become Hunters and resolve the match.
- **Simultaneous claims on the same resource:** Not applicable; turns are strictly sequential.
- **Player elimination mid-round:** When an agent is eliminated (`is_active` set to false), they are permanently skipped in the turn rotation.
- **Resource exhaustion:** Not applicable; marbles are persistent.
- **Termination guarantee:** Guaranteed by the expansion of the Mora radius (R6 gap-fill) combined with the non-zero probability of hitting static targets.
## State representation
- **Persistent state:**
- `agents`: Array of structs `(id, x, y, is_hunter, is_active, skill)`.
- `active_agent_index`: Integer.
- **Per-turn ephemeral state:**
- `intended_target`: Coordinate .
- `resolved_target`: Coordinate .
## Action space
- `flick_marble(target_x, target_y)`: Propel the marble toward target coordinates. Precondition: agent is active and it is their turn.
## Playability notes
- **Single-player-vs-AI:** Highly playable. The user controls a marble via an on-screen aiming reticle, while AI agents make decisions based on proximity, targeting the Mora if not a Hunter, or targeting the closest player if they are a Hunter.
- **Group games (n>4):** Scales naturally with no modification; sequential turn order expands.
- **Physical-skill games:** [gap-fill] Hand-flicking skill is modeled using a standard Gaussian error distribution around the target vector, with the variance parameter determined by the agent's dexterity level.
- **What is lost in the adaptation:** The physical tactile feedback of the glass marble, the friction and unevenness of the dirt surface, and the exact physical momentum transfer during collisions are abstracted into statistical equations.
## Implementation hints
- Implement the continuous 2D coordinate board state and a simple visual render.
- Write the distance collision check to resolve strikes and Mora entries.
- Implement the state machine that toggles Hunter status on entering the Mora and forces a recharge after an elimination.
- Develop a basic AI heuristic: "If not hunter, aim at (0,0); if hunter, aim at nearest active opponent."
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Marbles (Mora variant) (Israel). (n.d.). Games of Countries. https://gamesofcountries.com/games/il-israel/gullot
@misc{gamesofcountries-il-israel-gullot,
title = {Marbles (Mora variant) (Israel)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/il-israel/gullot}
}