Formal Specification
The Morra — Formal Specification
Italy · guessing-hidden · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: directGap-fills: 2
# La Morra — AI-optimized specification
## General idea
A simultaneous-choice zero-sum prediction game where two agents simultaneously output a finger count (1 to 5) and a total sum guess (2 to 10). A point is scored only when exactly one agent correctly predicts the combined sum of both players' revealed fingers. The core dynamic is psychological game theory and pattern recognition under simultaneous imperfect information.
## Formal model
- **Abstraction taken:** The acoustic and physical synchronization of throwing hands is abstracted into a simultaneous step where both agents commit private decisions before state reveal.
- **Player/agent count:** 2 active agents per round (in 4-player team matches, two non-active agents wait in a queue).
- **Turn model:** Simultaneous submission per round, followed by immediate deterministic resolution.
## Formal rules
R1. State setup: Each agent starts with a score of 0. Target score is (or ).
R2. In each round , both agents simultaneously and independently choose an action tuple , where is the fingers displayed and is the predicted total.
R3. The true total sum is calculated as .
R4. Point evaluation:
- If and , Agent 1 scores point.
- If and , Agent 2 scores point.
- If and (double correct), 0 points awarded.
- If and (double miss), 0 points awarded.
R5. Win condition: The first agent whose running score reaches wins the match immediately.
R6. [gap-fill] In rotational team mode (2 vs 2), when an active agent concedes a point, that agent transitions to the bench queue and their teammate becomes the active thrower for the next round.
R7. [gap-fill] A global round limit of 500 rounds is enforced; if neither agent reaches , the agent with the strictly higher score wins, and ties are broken by sudden-death single-point rounds.
## Edge cases
- Tie at the end condition: Impossible in standard 1v1 play because points are awarded strictly in increments of 1 to a single agent per round; simultaneous scoring awards 0 points.
- A player with no legal move: Not applicable; all action tuples are always valid every round.
- Repetition / cycles: Both agents may repeatedly guess incorrectly; bounded by R7 (move limit and sudden-death resolver).
- Simultaneous claims on the same resource or position: When both agents guess the identical true sum (), R4 explicitly voids the round with zero point allocation.
- Player elimination mid-round; turn order repair: In 2v2 team play, upon a point conceded, the conceding agent swaps with their teammate; turn order proceeds seamlessly with the newly active teammate.
- Resource exhaustion (pieces, tokens, space, time): Not applicable; actions are drawn from a non-depleting discrete set.
- Termination guarantee: Because each round has a non-zero probability of an asymmetric correct guess under mixed-strategy Nash equilibria (), the expected rounds to reach target score is finite.
## State representation
- Persistent state:
- `score`: array `[int, int]` tracking cumulative points for Agent 1 and Agent 2.
- `target_score`: integer (default 16).
- `round_count`: integer.
- `history`: list of past round tuples `[(f1, g1, f2, g2, S, point_awarded), ...]`.
- Per-round ephemeral state:
- `pending_action_1`: private tuple `(f1, g1)`.
- `pending_action_2`: private tuple `(f2, g2)`.
- Observable information: Both agents observe complete historical rounds; current round commitments are strictly hidden until simultaneous reveal.
## Action space
- `commit_throw(fingers: int, guess: int)`:
- Preconditions: , .
## Playability notes
- Single-player-vs-AI: Engine-ready; AI can implement reinforcement learning or Markov chain opponent modeling to exploit human finger/guess distribution biases.
- Group games (n>4): Naturally reduces to 2-agent duels or tournament elimination brackets.
- Physical-skill games: Direct strategic fidelity; physical coordination is abstracted to pure timing-neutral simultaneous disclosure.
- What is lost in the adaptation: The vocal intimidation, high-decibel dialect chants (*murreris*), and visual feints of physical play are omitted.
## Implementation hints
Extremely lightweight discrete state model. The payoff matrix per round is . Branching factor is 45 discrete actions per agent per round.
First milestone: Implement the simultaneous action resolver and state engine with uniform random bots, then integrate an n-gram frequency predictor.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
The Morra (Italy). (n.d.). Games of Countries. https://gamesofcountries.com/games/it-italy/la-morra
@misc{gamesofcountries-it-italy-la-morra,
title = {The Morra (Italy)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/it-italy/la-morra}
}