Formal Specification
Knucklebone shooting — Formal Specification
Mongolia · dexterity-small · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 3
# Shagai kharvakh — AI-optimized specification
## General idea
A team-based precision target game where two squads of shooters take turns flicking a polished bone striker from a stationary wooden guide toward an elevated rack of target knucklebones situated 4.72 metres away. Strategic depth lies in shot selection (targeting clustered vs. isolated bones) and team shooting order management, while the physical finger flick is modelled as a parameterized stochastic contest.
## Formal model
- **Abstraction taken:** The physical projectile flick (guide alignment, finger force, striker spin, bone collision dynamics) is abstracted into a parameterized probability distribution `P(hit | aim_target, shooter_skill)` that determines which subset of the remaining target bones are dislodged.
- **Player/agent count:** 2 teams of 6 agents each (12 agents total), shooting in fixed rotational order.
- **Turn model:** Sequential discrete turns across 4 designated shooting rounds (*üye*), each agent taking 4 shots per round.
## Formal rules
R1. Setup: 30 target knucklebones (*khasa*) are arranged on the target rack in two 15-bone rows. State variable `targets_remaining = 30`.
R2. Turn order: Agents shoot in fixed alternation between Team A and Team B (A1, B1, A2, B2, ..., A6, B6), repeating for 4 shots per agent per round.
R3. On each shot, the active agent selects an aim target coordinate on the target rack.
R4. Stochastic resolution: The shot outcome is sampled from `f(aim_target, shooter_skill, rack_state)`, returning `dislodged_count` (an integer in `[0, min(targets_remaining, max_possible)]`).
R5. State update: `targets_remaining` is decremented by `dislodged_count`, and the shooting team's cumulative score is incremented by `dislodged_count`.
R6. [gap-fill] If `targets_remaining == 0` before the round concludes, the rack is immediately replenished with 30 fresh targets to ensure subsequent shooters have targets available.
R7. Round completion: A match concludes after all 6 agents on both teams complete their prescribed shots across all 4 rounds (total 48 shots per team).
R8. Win condition: The team with the highest cumulative hit score at the end of the match wins.
R9. [gap-fill] Tie-breaking: If tied at regulation end, sudden-death single-shot rounds continue between nominated team captains until one team scores and the other misses.
## Edge cases
- Tie at the end condition: Resolved via R9 sudden-death single-shot captain shoot-offs.
- A player with no legal move: Not applicable — shooting an aim coordinate is always legal on every turn.
- Repetition / cycles: Not applicable — finite total turn count (48 shots per team) bounds the game provably.
- Simultaneous claims on the same resource: Not applicable — turns are strictly sequential.
- Player elimination mid-round: [gap-fill] If an agent is disqualified, a designated reserve agent takes their exact slot in the shooting rotation; team score is preserved.
- Resource exhaustion: Addressed by R6 rack replenishment rule if all 30 target bones are cleared mid-round.
- Termination guarantee: Provably terminates after exactly 96 total shots in regulation (plus finite geometric decay in sudden-death shoot-offs).
## State representation
- Persistent state: `team_scores` (map of Team -> integer), `round_index` (1..4), `shot_index` (1..4), `active_agent_index` (0..11), `targets_remaining` (integer, 0..30), `rack_layout` (30-element boolean array of present/dislodged tokens).
- Per-turn ephemeral state: Active agent ID, chosen aim target `(x, y)`, sampled `dislodged_count`.
- Observable information: Fully observable — all agent skill ratings, scores, shot histories, and target rack states are common knowledge.
## Action space
- `shoot(target_x, target_y)`: Select continuous or discretized coordinate on the target ledge. Precondition: agent is active in turn order.
## Playability notes
- Single-player-vs-AI: Natural — 1 human + 5 AI agents vs. 6 AI agents, or single-shooter target practice against baseline hit thresholds.
- Group games (n>4): Reduces cleanly to 1v1 by having 1 agent represent each team taking all 48 shots, preserving the strategic scoring mechanics while removing multi-agent coordination.
- Physical-skill games: Skill parameterized as scalar accuracy `σ_shooter`, where deviation from aim point follows a 2D Gaussian `N(0, σ^2)`.
- What is lost in the adaptation: The tactile sensation of the bone guide, hand tremors, acoustic feedback of colliding bones, and live team *uukhai* vocal synchronization.
## Implementation hints
Extremely low state complexity. Discretize the target rack into 30 distinct cell coordinates. Milestone 1: implement deterministic turn counter and scoring with a uniform 50% hit sampler; Milestone 2: implement Gaussian spatial aim sampler with cluster physics.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Knucklebone shooting (Mongolia). (n.d.). Games of Countries. https://gamesofcountries.com/games/mn-mongolia/shagai-kharvakh
@misc{gamesofcountries-mn-mongolia-shagai-kharvakh,
title = {Knucklebone shooting (Mongolia)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/mn-mongolia/shagai-kharvakh}
}