Formal Specification
Grey Wolf — Formal Specification
Kyrgyzstan · animal-contest · Confidence: high
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 2
# Kök börü — AI-optimized specification
## General idea
A high-intensity equestrian territorial contest where two teams of mounted riders vie for possession of a heavy goat carcass and maneuver across a rectangular pitch to throw it into an elevated goal cauldron (*tai-kazan*). It differs from conventional ball team sports due to the high mass and friction of the target object, the dynamic body-checking physics of horses, and the continuous one-on-one horse wrestling required to extract the carcass from under opponents' saddle flanks.
## Formal model
- **Abstraction taken:** The continuous 3D physics of horseback riding, galloping inertia, and torso-grappling are discretized into a 2D zoned pitch (e.g. 8x4 grid) with discrete time ticks (e.g. 1 second per tick). Physical carcass grappling and snatching are modeled as contested probabilistic skill checks parameterized by rider strength, horse mass, and positioning.
- **Player/agent count:** 2 teams, 4 active agents per team on field.
- **Turn model:** Simultaneous discrete ticks; agents submit intentions concurrently, resolved in a deterministic priority order.
## Formal rules
R1. Setup: Field divided into an 8x4 grid with team zones, center circle at (4,2)-(5,3), and two *tai-kazan* goal cauldrons at grid cells (1,2)-(1,3) and (8,2)-(8,3). Carcass initialized at center point in neutral unpossessed state.
R2. Per-tick actions: Agents simultaneously declare one action from `move(direction)`, `snatch_carcass()`, `grapple_possession()`, `pass_carcass(target_agent)`, `shoot_goal()`, or `screen_opponent(target_agent)`.
R3. Possession mechanics:
- If carcass is free on the ground in cell (x,y), an agent in (x,y) attempting `snatch_carcass()` succeeds with probability `P(snatch) = f(rider_agility, horse_stability)`.
- If carcass is held by an agent in (x,y), an adjacent opposing agent executing `grapple_possession()` triggers a contest resolver: `P(steal) = f(attacker_str, defender_str, horse_mass_diff)`.
R4. Goal scoring: An agent in a goal-adjacent cell executing `shoot_goal()` scores a goal (*салым*) if `P(goal) = f(distance, defensive_pressure)` succeeds. Upon a goal, attacking team receives +1 point, all agents reset to their defensive halves, and the carcass resets to center.
R5. Match structure: Game lasts exactly 3 periods of 120 ticks each (representing 3 periods of 20 minutes).
R6. Win/loss: The team with the higher goal tally at tick 360 wins. [gap-fill] If tied at tick 360, a 60-tick sudden-death overtime is played; if still tied, a 1v1 shootout resolves the match.
## Edge cases
- Tie at the end condition: Resolved via 1v1 sudden-death shootout where one rider from each team races from midfield to score.
- A player with no legal move: Not applicable; `move(stay)` and defensive posture actions are always available.
- Repetition / cycles: Bounded by fixed match duration (360 ticks). [gap-fill] A 30-tick shot-clock rule forces a turnover if a possessing team does not advance the carcass into the attacking half.
- Simultaneous claims on the same resource: If two opposing agents execute `snatch_carcass()` on a free carcass on the exact same tick, the contest resolver compares `strength + agility` modifiers with tied rolls leaving the carcass contested on the ground.
- Player elimination mid-round: If an agent incurs a 2-minute penalty (12 ticks in penalty box), their team plays with 3 agents until the penalty expires.
- Resource exhaustion: Carcass never leaves play; if deflected out of bounds, it respawns at the nearest boundary hash mark.
- Termination guarantee: Provably terminates at tick 360 (or tick 420 in overtime).
## State representation
- Persistent state: Score per team (int), period (1..3), match tick (0..360), penalty box timers per agent.
- Agent state: 2D grid position (x,y), velocity vector, possession status (boolean), stamina (float 0..100), strength attribute, horsemanship attribute.
- Ephemeral state: Carcass position (x,y) or carrier ID, declared simultaneous actions.
- Observability: Fully observable by all agents.
## Action space
- `move(dx, dy)`: Advance 1 cell in cardinal/diagonal direction (precondition: within pitch bounds).
- `snatch_carcass()`: Attempt ground pickup (precondition: carcass is free in current cell).
- `grapple_possession(target_agent)`: Contest carcass held by adjacent opponent.
- `pass_carcass(target_agent)`: Pass to teammate within 2 cells.
- `shoot_goal()`: Cast carcass into opposing *tai-kazan* (precondition: carrier is within 1 cell of goal).
## Playability notes
- Single-player-vs-AI: Natural; player controls 1 rider while teammate/opponent agents execute coordinate behavioral trees or RL policies.
- Group games (n=8): Easily reduced to 2 human captains commanding squads of 4 agents.
- Physical skill model: Equestrian momentum and carcass wrestling modeled via stamina depletion and strength/mass differentials.
- What is lost: The tactile resistance of lifting a wet 35 kg carcass at 40 km/h and fine rein-handling are replaced by discrete probability rolls.
## Implementation hints
Branching factor is roughly 6 actions per agent across 8 agents (~1.6 million joint action combinations per tick), which can be pruned by resolving agents sequentially based on speed/initiative order. First milestone: implement a 2D grid simulation with 2v2 agents, deterministic carcass picking, and goal scoring before adding grappling probability modifiers.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Grey Wolf (Kyrgyzstan). (n.d.). Games of Countries. https://gamesofcountries.com/games/kg-kyrgyzstan/kok-boru
@misc{gamesofcountries-kg-kyrgyzstan-kok-boru,
title = {Grey Wolf (Kyrgyzstan)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/kg-kyrgyzstan/kok-boru}
}