Skip to main content
Formal Specification

Knucklebones — Formal Specification

Bulgaria · dexterity-small · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 3
# Ashitsi — AI-optimized specification ## General idea A turn-based staking and trajectory target game played with anatomically asymmetric four-sided astragalus bones. Players commit stakes to a central target layout and execute parameterized flick/throw shots to dislodge target pieces, with turn continuation granted upon successful captures and bonus scoring determined by the landed bone face distribution. ## Formal model - **Abstraction taken:** The physical roll and collision dynamics of sheep knucklebones are abstracted into: (1) an asymmetric 4-state probability distribution for free-landing bones (PDyup,PGarb,PChif,PTek), and (2) a 2D collision vector model for shooting target displacement. - **Player/agent count:** 2 to 6 players (K∈[2,6]). - **Turn model:** Sequential turn rotation with strike-continuation on successful hit. ## Formal rules R1. Astragalus randomizer model: When an astragalus is tossed freely, it lands in one of 4 discrete states with non-uniform probability: - State 0 (*Dyup* / Concave): p0=0.40, Value = 1 pt. - State 1 (*Garb* / Convex): p1=0.35, Value = 2 pts. - State 2 (*Chif* / Flat): p2=0.15, Value = 5 pts. - State 3 (*Tek* / Sinuous): p3=0.10, Value = 10 pts. R2. Setup: Each player stakes S bones (S≥1). Total pot M=K×S bones are placed at positions xi=(i⋅Δd,0) for i∈[0,M−1] along a line at distance Ytarget=4.0 metres. R3. Turn order determination: Each agent submits an initial aim vector toward line Ydist=3.0; agent with smallest landing error shoots first in sequence. R4. Shooting turn: - Active agent submits aim vector 𝐯=(θ,F) from throwing line y=0. - [gap-fill] Shooter landing position 𝐩biyach=(Fsin⁡θ,Fcos⁡θ)+𝒩(0,Σskill). - Collision check: For each remaining target bone j at position 𝐩j, if ||𝐩biyach−𝐩j||<rhit, bone j is struck. - Struck bone displacement: Δ𝐩j=𝐯transfer. If ||Δ𝐩j||>dcapture_threshold, bone j is captured and added to the active agent's bank. R5. Turn continuation rule: - If at least 1 bone is captured during the throw, active agent retains the turn and throws again from current position 𝐩biyach. - If 0 bones are captured, the turn ends immediately; turn passes to (current_player+1)(modK). R6. [gap-fill] Maximum turn limit: An active player may execute at most 4 consecutive scoring shots in a single turn to prevent full table-clearing runouts. R7. Victory: Game ends when all M target bones have been captured. The player with the highest count of captured bones wins the round. R8. [gap-fill] Tiebreaker: If multiple players tie for most captured bones, each tied player tosses 1 bone; highest face value (Tek>Chif>Garb>Dyup) breaks the tie. ## Edge cases - Tie at the end condition: Fully resolved by R8 sudden-death bone toss. - A player with no legal move: Not applicable — shooting action with aim vector is always legal while target bones remain. - Repetition / cycles: Bounded: The number of remaining target bones M is strictly non-increasing and decreases by at least 1 on each successful shot, terminating when M=0. - Simultaneous claims on the same resource or position: Turns are strictly sequential; exact bone objects are tracked and removed from play upon capture. - Player elimination mid-round; turn order repair: If a player disconnects, their captured bones remain banked; turn skips disconnected player ID in sequence. - Resource exhaustion: Target bones decrement to exactly 0, triggering game end condition R7. - Termination guarantee: Enforced by finite pool M and mandatory turn pass on miss (R5). ## State representation - Persistent state: Player banks [B1,…,BK], active player index c∈[0,K−1], shooter positions 𝐩biyach_i. - Per-turn ephemeral state: Remaining target positions array [𝐩1,…,𝐩m], consecutive hit counter, active shot vector. - Observable information: Fully observable — all bone positions and bank counts are public knowledge. ## Action space - `shoot_target(angle, force)`: Precondition: Caller is active player, target bones remain (m>0). - `toss_bone_face()`: Evaluates the 4-state asymmetric randomizer for tiebreaking or face-matching variants. ## Playability notes - Single-player-vs-AI: Natural — Human plays vs 1..5 AI opponents with parameterized shot accuracy and aim selection logic. - Group games (n>4): Fully supported up to 6 players natively without structural changes. - Physical-skill games: Accuracy modeled via bivariate normal distribution centered on intended target, with variance inversely proportional to agent `dexterity` attribute. - What is lost: The physical hand feel of selecting smooth versus lead-filled astragali and manual wrist flick technique is parameterized. ## Implementation hints Low complexity. State is a small list of 2D coordinates and player bone counts. The asymmetric discrete distribution R1 provides instant authenticity for any dice/face-rolling mechanics. Suggested first milestone: implement 1D line-shooting simulation with uniform target spacing and verify termination when all targets are cleared.

Download as JSON

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

Download JSON
Cite this page

Knucklebones (Bulgaria). (n.d.). Games of Countries. https://gamesofcountries.com/games/bg-bulgaria/ashitsi

@misc{gamesofcountries-bg-bulgaria-ashitsi,
  title = {Knucklebones (Bulgaria)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/bg-bulgaria/ashitsi}
}