Skip to main content
Formal Specification

Jacks — Formal Specification

Brazil · dexterity-small · Confidence: high

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 3
# Cinco Marias — AI-optimized specification ## General idea Five-stone jacks, structured as a strict difficulty ladder: throw one piece, gather a fixed number from the ground, catch the thrown piece before it lands. The number gathered per throw rises each level. It is a solitaire skill sequence played competitively by comparison, with no interaction between players at all. ## Formal model - **Abstraction taken:** the throw-and-catch is replaced by a single success probability per attempt, rising in difficulty with the number of pieces to be gathered. No trajectory or hand motion is modelled. - **Player/agent count:** 1–4 or more, but they never interact; the multiplayer form is parallel solitaire with turn passing on failure. - **Turn model:** sequential; a player retains the turn until a drop. ## Formal rules R1. Five pieces begin on the ground. Play proceeds through levels L=1,2,3,4. R2. At level L, an attempt is: throw one piece up, gather exactly L pieces from the ground, catch the thrown piece in the same hand. R3. An attempt succeeds with probability `p(L, skill)`, strictly decreasing in L. [gap-fill] No source gives throw height, air time, or catching requirements — the entry marks all three undocumented — so the entire physical act is one parameter. R4. Success advances the sequence; the gathered pieces are set aside and the level continues until all five have been gathered, then L increases. R5. A drop ends the turn immediately, and play passes to the next player. [gap-fill] The failed level is resumed on the player's next turn rather than restarted; sources do not specify, and restarting would make the ladder unwinnable for most players. R6. [gap-fill] Levels beyond L=4 exist in play but are not documented. The spec stops at L=4 and treats completion of L=4 as the win. R7. The first player to complete level 4 wins. ## Edge cases - **Tie at the end condition:** impossible during play, since turns are sequential and one player completes L=4 strictly before another. If a fixed number of rounds is imposed instead, ties are broken by highest level reached, then fewest attempts. - **Player with no legal move:** cannot occur. A player on turn always has an attempt available; a player not on turn simply waits. - **Repetition / cycles:** the level index never decreases (R5), so the state strictly progresses and no cycle exists. - **Simultaneous claims:** none. Only one player touches the pieces at a time and there is no shared contested resource. - **Mid-round elimination:** there is no elimination. Turn order is a fixed rotation and needs no repair. - **Resource exhaustion:** the five pieces are conserved; nothing is consumed. - **Termination guarantee:** each attempt succeeds with probability bounded below by a positive constant, so every level is cleared in finite expected attempts and the ladder has four levels. The game ends with probability 1. ## State representation Persistent: piece count (5), the level ladder L=1..4. Per-player: current level, pieces gathered at the current level, skill parameter. Per-turn ephemeral: whether the current attempt succeeded. Information set: fully observable, and irrelevant — no player's choice depends on another's state. ## Action space - `attempt(level)` — precondition: it is the actor's turn and the actor's current level is `level`. This is the only action in the game. ## Playability notes - **Single-player-vs-AI:** the game is *already* single-player. There is no opponent model to build and no adaptation needed; an agent has no choices, only a skill parameter. - **Group games (n>2):** reduction to two agents is trivial and lossless, because players never interact. This is worth stating plainly: the multiplayer form adds social stakes, not game structure. - **Physical-skill model:** one probability per level, decreasing in L. That is the entire model, and it is also the entire game. - **What is lost:** the game has no strategic content to lose — but the *dexterity* is the whole point, and reducing it to a probability leaves an artifact that can be scored and not played. This is close to the boundary of `non-simulable`; it stays `probabilistic-proxy` only because the level ladder gives it discrete structure. ## Implementation hints There is no search here and no branching factor to speak of — the action space has size one. Implement it as a Markov chain over (level, pieces gathered) and it is complete. If you want a game rather than a simulation, the interesting work is in modelling the skill parameter's improvement over repeated play, not in the rules.

Download as JSON

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

Download JSON
Cite this page

Jacks (Brazil). (n.d.). Games of Countries. https://gamesofcountries.com/games/br-brazil/cinco-marias

@misc{gamesofcountries-br-brazil-cinco-marias,
  title = {Jacks (Brazil)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/br-brazil/cinco-marias}
}