Skip to main content
Formal Specification

King, Minister, Thief, Soldier — Formal Specification

Nepal · guessing-hidden · Confidence: low

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: directGap-fills: 0
# Raja Mantri Chor Sipahi — AI-optimized specification ## General idea Raja Mantri Chor Sipahi is a 4-player discrete social deduction and hidden role guessing game. 4 unique role chits (King 1000, Minister 800, Soldier 500, Thief 0) are secretly distributed. The King and Minister reveal themselves publicly, and the Soldier must interrogate the remaining two hidden agents to deduce which one is the Thief and which is the Minister/Thief counter-role. ## Formal model - **Abstraction taken:** Social facial tells are formalized as a probabilistic observation model: during interrogation, the Thief emits a deception feature vector X∼𝒩(μthief,Σ) while the innocent Minister emits X∼𝒩(μminister,Σ). - **Player/agent count:** Exactly 4 agents (Agent 0, 1, 2, 3). - **Turn model:** Round-based; each round has a Deal Phase, Reveal Phase, Interrogation Phase, and Resolution Phase. ## Formal rules R1. Role Deck: Multiset of 4 roles ℛ={KING: 1000,MINISTER: 800,SOLDIER: 500,THIEF: 0}. R2. Deal Phase: Roles are randomly permuted uniformly across the 4 agents: π:{0,1,2,3}→ℛ. Each agent observes only their own assigned role π(i). R3. Public Announcement Phase: - Agent holding KING declares identity publicly →Publicly Known. - Agent holding MINISTER reveals role →Publicly Known. - Agent holding SOLDIER reveals role →Publicly Known. - The remaining two unrevealed agents hold {MINISTER,THIEF} (or King's variant). In the standard canonical protocol, King and Soldier are public; Minister and Thief are hidden. R4. Accusation / Deduction Phase: - The Soldier agent inspects the two candidate agents {Ax,Ay}. - Soldier executes action `accuse_thief(target_agent_id)`. R5. Scoring Resolution: - If accused agent ==THIEF (Correct Deduction): - King receives 1000 pts. - Minister receives 800 pts. - Soldier receives 500 pts. - Thief receives 0 pts. - If accused agent ==MINISTER (Wrong Deduction / False Accusation): - King receives 1000 pts. - Minister receives 800 pts. - Thief STEALS Soldier's points: Thief receives 500 pts, Soldier receives 0 pts. R6. Match Termination: Match consists of M=12 rounds (each agent holds each role 3 times in expectation). R7. Win Condition: The agent with the highest cumulative point total after M rounds wins. ## Edge cases - Tie at the end condition: Ties in cumulative score broken by highest count of successful Thief deductions as Soldier. - A player with no legal move: Soldier must select exactly 1 of the 2 candidate suspects. - Repetition / cycles: Bounded strictly by finite M=12 rounds. - Simultaneous claims on the same resource or position: Actions are strictly sequential within the phase. - Player elimination mid-round: Non-elimination party game. - Resource exhaustion: Exactly 4 chits dealt per round. - Termination guarantee: Provably terminates after round M. ## State representation - `role_assignment`: Private array of 4 roles. - `public_roles`: Array of 4 visible role enums ∈{UNKNOWN,KING,MINISTER,SOLDIER,THIEF}. - `cumulative_scores`: Array of 4 integers [S0,S1,S2,S3]. - `round_number`: Integer ∈[1,12]. ## Action space - `reveal_role()`: Publicly confirm identity during proclamation phase. - `accuse_thief(target_agent_id)`: Valid only for Soldier agent on unrevealed agents. ## Playability notes - Single-player-vs-AI: Fully playable against 3 AI agents running Bayesian deception classifiers. - Group games (n>4): Strictly fixed at 4 players (standard parlor format). - Physical-skill games: Facial reading and bluffing modeled as Bayesian signal detection. - What is lost in the adaptation: The theatrical comedic acting of innocent players pleading before the Soldier. ## Implementation hints Ideal benchmark for hidden-information Bayesian inference and imperfect-information game theory. State space is tiny (4!=24 permutations per round). First milestone: simulate 1,000 matches with random vs. Bayesian Soldier agents.

Download as JSON

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

Download JSON
Cite this page

King, Minister, Thief, Soldier (Nepal). (n.d.). Games of Countries. https://gamesofcountries.com/games/np-nepal/raja-mantri-chor-sipahi

@misc{gamesofcountries-np-nepal-raja-mantri-chor-sipahi,
  title = {King, Minister, Thief, Soldier (Nepal)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/np-nepal/raja-mantri-chor-sipahi}
}