Formal Specification
Knucklebones — Formal Specification
Afghanistan · dexterity-small · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 4
# Bujul-bazi — AI-optimized specification
## General idea
A sequential target-shooting and collection game where agents use a weighted shooter knucklebone (*sakha*) to dislodge sheep astragalus bones from a drawn target circle. Successfully knocking knucklebones out of the ring claims them as permanent victory points and earns consecutive bonus shots, while faults (failing to dislodge any bone or stranding the shooter bone inside the ring) immediately forfeit the turn.
## Formal model
- **Abstraction taken:** Continuous physics of knucklebone mass, rotation, elastic collisions, and ground friction is abstracted into a discrete 2D spatial layout where shots specify an aim angle and force vector, evaluated by a stochastic collision resolver.
- **Player/agent count:** 2 to 6 individual agents.
- **Turn model:** Sequential turns with variable length (successful captures grant re-shots until a miss or fault occurs).
## Formal rules
R1. **Setup:** Each of the participating agents contributes knucklebones () to form an ante pool of bones clustered within a circle of radius meter centered at . Each bone is initialized at a polar coordinate with and an anatomical resting state `side` .
R2. **Turn Order Determination:** At game start, each agent executes a preliminary toss. Turn order is sorted descending by bone orientation rank (), breaking ties by distance to circle center. [gap-fill: distance tie-breaker formalized].
R3. **Turn Execution:** The active agent stands at the throwing boundary with meters and chooses an action `shoot(angle, power)`.
R4. **Shot Resolution:**
- The shooter trajectory and collision dynamics are sampled via a physical proxy model: .
- Every knucklebone whose resulting position is removed from the ring and added to the active agent's captured stash.
- The shooter bone comes to rest at coordinate .
R5. **Bonus Shot & Turn Transition:**
- If at least one knucklebone was dislodged () AND the shooter bone came to rest outside the ring (), the active agent retains the turn and shoots again from .
- If no knucklebone was dislodged, or if the shooter bone came to rest inside the ring (*sakha sokhtan* fault), the turn ends immediately. [gap-fill: If *sakha sokhtan* occurs, the shooter bone is retrieved to the boundary line with no penalty other than immediate turn forfeiture].
- On turn end, play passes to the next active agent in the cyclic order.
R6. **Round End & Victory:** The game terminates immediately when the count of remaining knucklebones in the circle reaches 0. The agent with the highest count of captured knucklebones wins. [gap-fill: If the game reaches 10 full rotation cycles without any bone captured, the round terminates and uncaptured bones are returned to the pool].
## Edge cases
- Tie at the end condition: If two or more agents have an equal number of captured knucklebones when the ring is emptied, the tied agents execute a sudden-death single toss; highest-ranking landing orientation wins. [gap-fill: sudden-death orientation toss for endgame ties].
- A player with no legal move: Not applicable; an active agent always has a legal `shoot(angle, power)` action.
- Repetition / cycles: Knucklebones in the ring are monotonically non-increasing (bones only leave the ring, never re-enter). If no bones are captured for 10 consecutive full rounds, the termination safety rule in R6 breaks any potential infinite miss cycle.
- Simultaneous claims on the same resource or position: Not applicable; turns and shots are strictly sequential.
- Player elimination mid-round; turn order repair: In standard play, all agents remain active in turn order until all bones are claimed. If an agent withdraws, they are removed from the rotation list and the next agent in cyclic order takes the turn.
- Resource exhaustion (pieces, tokens, space, time): The pool of knucklebones is strictly finite and decrements upon each successful capture until reaching exactly 0.
- Termination guarantee: Provably terminates because the target bone pool is finite () and strictly decreases toward 0, bounded from infinite misses by the 10-round cycle timeout rule.
## State representation
- **Persistent state:**
- `agents`: list of agent IDs in cyclic turn order
- `captured_counts`: map of `agent_id -> integer`
- `knucklebones_in_ring`: list of `{id, x, y, side}`
- `active_turn_index`: integer `0..N-1`
- `shooter_pos`: `{x, y}` (current position of active shooter bone)
- `consecutive_miss_rounds`: integer
- **Per-turn ephemeral state:**
- `chosen_angle`: float ()
- `chosen_power`: float ()
- `bones_knocked_out`: list of bone IDs
- `fault_occurred`: boolean
- **Observable information:** Fully observable. All bone positions, shooter position, circle radius, and score tallies are public.
## Action space
- `shoot(angle, power)`:
- `angle`: float in radians relative to target center.
- `power`: float in range .
- Preconditions: Agent is the active player and `len(knucklebones_in_ring) > 0`.
## Playability notes
- Single-player-vs-AI: Highly playable. A single human player can compete against 1 to 5 AI agents using geometric heuristic target selection (aiming at clustered or perimeter bones).
- Group games (n>4): Reduces seamlessly to 2 agents (head-to-head match with bones) without loss of tactical depth or turn-taking dynamics.
- Physical-skill games: Physical dexterity (flicking trajectory and wrist spin) is mapped to an error distribution parameterized by agent skill: .
- What is lost in the adaptation: The tactile feel of bone heft, lead-drilled balance customization, and varying friction of outdoor dirt versus indoor felt mats are simplified into numerical collision mechanics.
## Implementation hints
Low-to-medium complexity. The geometric collision model can be implemented as 2D circle-circle elastic collision physics or simplified to ray-cast impact tests with momentum transfer. Suggested first milestone: implement turn management and discrete bone removal with a deterministic hit box, then layer the stochastic angle/power variance model.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Knucklebones (Afghanistan). (n.d.). Games of Countries. https://gamesofcountries.com/games/af-afghanistan/bujul-bazi
@misc{gamesofcountries-af-afghanistan-bujul-bazi,
title = {Knucklebones (Afghanistan)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/af-afghanistan/bujul-bazi}
}