Skip to main content
Formal Specification

Breaking the Clay Pot — Formal Specification

Andorra · festival-ritual · Confidence: low

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 2
# Trencar l'olla — AI-optimized specification ## General idea A turn-based blindfolded navigation and targeting game where an active agent, disoriented by initial rotation noise, moves across a 2D plane attempting to strike a hidden target pot using noisy directional signals (a mixture of truthful and deceptive crowd guidance). ## Formal model - **Abstraction taken:** The 3D physical swing and blindfolded movement are abstracted into a 2D continuous coordinates `(x, y)` and heading angle `theta`. The acoustic crowd shouting is abstracted into a Bayesian observation model with signal-to-noise ratio `SNR`. - **Player/agent count:** 1 to N sequential agents. - **Turn model:** Strictly sequential single-agent trials. ## Formal rules R1. Setup: Target pot is positioned at hidden true location `(X_pot, Y_pot)` within a bounded radius `R_arena = 5.0` meters. R2. Initialization: The active agent starts at `(0, 0)` with heading `theta ~ Uniform(0, 2*pi)` (modeling the initial spin disorientation). Swing counter `swings_left = 3`. R3. On each step `t`: R3.1. The agent receives a noisy directional hint `alpha_obs = angle_to(X_pot, Y_pot) + noise`, where `noise ~ Normal(0, sigma_crowd)` with probability `p_truth` (truthful guidance), or `noise ~ Uniform(-pi, pi)` with probability `1 - p_truth` (deceptive hecklers). R3.2. The agent selects an action: `step(distance, turn_angle)` or `swing()`. R3.3. If `swing()` is executed: `swings_left` decrements by 1. R3.4. A swing succeeds if Euclidean distance `d(pos, (X_pot, Y_pot)) <= R_hit` (default `R_hit = 0.8` meters) and `|heading - angle_to(pot)| <= theta_hit` (default `pi/4`). R3.5. If the swing succeeds, the pot is shattered (`pot_broken = true`), the agent scores 100 points minus step count penalty, and the trial ends. R3.6. If `swings_left == 0` and `pot_broken == false`, the trial ends with 0 points. R4. [gap-fill] Boundary bounce: If an agent steps outside `R_arena`, their position is clamped to the boundary and an audible buzzer warning is returned. R5. Game terminates when all registered agents have completed one trial. The agent with the highest score wins. ## Edge cases - Tie at the end condition: Resolved by fewest steps taken to smash the pot. - A player with no legal move: Not applicable — stepping or swinging is always legal while `swings_left > 0`. - Repetition / cycles: Bounded strictly by `swings_left = 3` and a maximum step cap `max_steps = 30` [gap-fill]. - Simultaneous claims on the same resource or position: Not applicable — trials are strictly sequential 1-agent turns. - Player elimination mid-round: Disconnection forfeits the remaining swings with score 0. - Resource exhaustion: If all pots are smashed in a multi-pot tournament, new pots are initialized at fresh random coordinates. - Termination guarantee: Guaranteed by the finite swing limit (3) and step limit (30). ## State representation - Persistent state: `leaderboard` (map of agent_id -> score). - Per-trial ephemeral state: `agent_pos` `(x, y)`, `agent_heading` `theta`, `pot_pos` `(X, Y)`, `swings_left` (int 0..3), `step_count` (int 0..30). - Observable information: Partially observable — `pot_pos` is strictly hidden from the agent. The agent observes only noisy guidance `alpha_obs`, wall bump warnings, and swing outcomes. ## Action space - `step(dist, angle)`: Move up to 1.0 meter and adjust heading. - `swing()`: Attempt to strike the overhead pot with the wooden stick. ## Playability notes - Single-player-vs-AI: Natural — the AI implements a particle filter or Kalman filter to estimate the hidden pot coordinates from contradictory noisy hints. - Group games (n>4): Sequential tournament format with shared leaderboard. - Physical-skill games: Stick swing reach and crowd acoustic noise parameterized via `R_hit` and `sigma_crowd`. - What is lost: The physical sensation of vertigo, tactile stick vibration, and comedic visual slapstick. ## Implementation hints Ideal demonstration of POMDP (Partially Observable Markov Decision Process) and belief-state filtering. Suggested milestone: Verify tracking accuracy under 70% truthful crowd signal before tuning swing hitboxes.

Download as JSON

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

Download JSON
Cite this page

Breaking the Clay Pot (Andorra). (n.d.). Games of Countries. https://gamesofcountries.com/games/ad-andorra/trencar-l-olla

@misc{gamesofcountries-ad-andorra-trencar-l-olla,
  title = {Breaking the Clay Pot (Andorra)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/ad-andorra/trencar-l-olla}
}