Skip to main content
Formal Specification

Italian Bowls (Raffa specialty) — Formal Specification

Italy · dexterity-throw · Confidence: low

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 2
# Bocce (specialità Raffa) — AI-optimized specification ## General idea A turn-based positional and target-displacement bowling game played on a bounded 2D rectangular lane. Agents alternate delivering balls to maximize the number of friendly balls closer to a target ball (*pallino*) than the closest opponent ball, utilizing rolling precision (*accosto*) or declared ballistic impacts (*raffa* / *volo*) that dynamically reconfigure the lane layout. ## Formal model - **Abstraction taken:** Continuous physical bowling aerodynamics and collision mechanics are modeled via a 2D coordinate system on a 28.0m×4.0m plane. Shot outcomes (final coordinates and elastic collisions) are generated by a physics-proxy distribution parameterized by the agent's chosen target, shot type, and execution skill variance. - **Player/agent count:** 2 opposing agents (or 2 teams of 2 to 3 agents). - **Turn model:** Dynamic sequential turns; the team not currently holding the closest ball to the pallino must throw next until they gain the point or exhaust their balls. ## Formal rules R1. Game setup: Target score T=12. A coin toss determines Agent A (starting) and Agent B. R2. Frame start: Active agent delivers the pallino, setting its coordinates (xp,yp) within the legal zone x∈[14.0,24.0],y∈[0.5,3.5]. R3. The active agent delivers their first boccia (rolling *accosto*). R4. Turn selection logic: The agent/team whose closest ball to the pallino is further than the opponent's closest ball must deliver the next ball. If an agent runs out of balls, the other agent throws all remaining balls. R5. Action execution: The agent chooses an action type ∈{accosto,raffa,volo} and a target coordinate or entity ID. - [gap-fill] *Accosto* lands at target (x,y)+𝒩(0,σacc2). - [gap-fill] *Raffa* / *Volo* requires a declared target entity; if hit, standard 2D rigid-body momentum transfer applies; if missed and illegal, the non-throwing agent chooses whether to apply the advantage rule (resetting disturbed items). R6. Frame scoring: When all 8 balls (4 per side in 1v1) are played, the side holding the ball at minimum Euclidean distance dmin⁡=mini⁡‖𝐛i−𝐩‖ scores k points, where k is the count of its balls closer to 𝐩 than any opponent ball. R7. Match termination: If an agent's score reaches T=12, that agent wins immediately. Otherwise, the winning team of the frame throws the pallino to start the next frame. ## Edge cases - Tie at the end condition: Frame ties (equidistant balls) award 0 points to both sides, and the pallino is redelivered; cumulative score ties at 11-11 proceed to an additional deciding frame. - A player with no legal move: Not applicable; an agent with remaining balls always has legal throw actions. - Repetition / cycles: Impossible because each frame strictly consumes all 8 finite balls in a maximum of 8 turns, and each frame yields non-negative score progression toward T=12. - Simultaneous claims on the same resource or position: Physical ball overlap is prevented by rigid circular collision radius (r=0.0535m); simultaneous turns do not exist. - Player elimination mid-round; turn order repair: In team play, if a player is disqualified or injured, remaining teammates absorb remaining deliveries without altering frame turn selection logic. - Resource exhaustion (pieces, tokens, space, time): Each frame exactly terminates when both sides have delivered all 4 balls; balls are reset each frame. - Termination guarantee: Since in each non-tied frame at least 1 point is awarded to the winning side, the game provably terminates in a finite number of frames (bounded above with high probability). ## State representation - Persistent state: - `score_a`, `score_b`: integers ∈[0,12]. - `frame_number`: integer. - Per-frame state: - `pallino_pos`: 2D vector (x,y). - `balls_on_court`: list of active balls with team ID, position (x,y), and status. - `balls_remaining_a`, `balls_remaining_b`: integers ∈[0,4]. - Observable information: Fully observable 2D board state to all agents. ## Action space - `throw_accosto(target_x: float, target_y: float)` - `declare_and_throw_raffa(target_id: int)` - `declare_and_throw_volo(target_id: int)` - `resolve_advantage(accept_layout: bool)` (invoked on illegal opponent shot) ## Playability notes - Single-player-vs-AI: Natural and engine-ready; AI evaluates spatial target density and selects optimal risk-adjusted shot type (rolling vs clearing). - Group games (n>4): Cleanly models doubles (2v2) or triples (3v3) by partitioning ball allocations across agents. - Physical-skill games: Skill is cleanly represented by Gaussian error distributions σ2 and declared shot collision probabilities. - What is lost in the adaptation: Fine surface irregularities, spinning ball slice (*sottomano* / *sopramano* spin), and tactile court reading are discretized into parameter variables. ## Implementation hints Discrete turn manager on top of a 2D circle-packing collision model. Branching factor is low if discretized into high-level tactical intents: Accosto-to-Pallino, Block-Path, Raffa-Opponent-Ball, Volo-Pallino. First milestone: Build the frame turn-evaluation logic and scoring comparator with exact Euclidean distance calculations before tuning the collision physics proxy.

Download as JSON

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

Download JSON
Cite this page

Italian Bowls (Raffa specialty) (Italy). (n.d.). Games of Countries. https://gamesofcountries.com/games/it-italy/bocce-raffa

@misc{gamesofcountries-it-italy-bocce-raffa,
  title = {Italian Bowls (Raffa specialty) (Italy)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/it-italy/bocce-raffa}
}