Formal Specification
The Thief and the Judge (Turkmen Role-Play Deduction Game) — Formal Specification
Turkmenistan · guessing-hidden · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 4
# Ogry bilen kazy — AI-optimized specification
## General idea
A traditional Turkmen asymmetric social deduction and interrogation parlor game. Hidden roles (Judge, Guard, Thief, Citizens) are secretly distributed; the Judge and Guard publicly reveal themselves, and the Guard must interrogate the unrevealed assembly to unmask the hidden Thief. A successful accusation convicts the Thief, while a false accusation convicts the Guard for incompetence.
## Formal model
- **Abstraction taken:** Non-verbal facial expressions, nervous fidgeting, and spoken courtroom dialogues are abstracted into a noisy signal emission and belief update framework. The Thief emits a deceptive signal vector, Citizens emit innocent baseline signals, and the Guard calculates posterior probabilities to make a single final accusation.
- **Player/agent count:** 4 to 12 agents.
- **Turn model:** Strictly sequential phases per round (Dealing Public Reveal Interrogation Accusation Sentencing).
## Formal rules
R1. **Role Allocation:** At round start, role cards are shuffled and dealt secretly: exactly 1 `JUDGE` (*Kazy*), 1 `GUARD` (*Sakçy*), 1 `THIEF` (*Ogry*), and `CITIZEN` (*Dogry / Raýat*).
R2. **Public Reveal Phase:**
- The `JUDGE` agent publicly reveals role and announces court is in session.
- The `GUARD` agent publicly reveals role upon the Judge's summons.
- The `THIEF` and `CITIZEN` agents keep their role assignments strictly hidden.
R3. **Interrogation & Signal Generation Phase:**
- The Guard queries each unrevealed agent .
- [gap-fill] Each queried agent emits a behavioral signal , where:
- For Citizens: (baseline innocence).
- For the Thief: , with modulated by the Thief's composure skill.
R4. **Accusation (*Ogryny görkezmek*):**
- The Guard selects exactly one unrevealed suspect and submits `Accuse(k)`.
- Suspect reveals their secret role card.
R5. **Judicial Resolution & Scoring:**
- **Case 1 (Accused is THIEF):**
- The Guard wins the round (+10 points to Guard, +5 to Judge, -10 to Thief).
- The Thief is sentenced to perform a penalty task.
- **Case 2 (Accused is CITIZEN):**
- The Guard loses the round (Wrongful Arrest: -10 points to Guard, +15 to Thief).
- The Guard is sentenced for incompetence; the Thief escapes unpunished.
R6. **Match Progression & Victory:**
- [gap-fill] Roles are collected, reshuffled, and re-dealt across rounds.
- [gap-fill] The player with the highest cumulative score across all role rotations wins the game.
## Edge cases
- Tie at the end condition: Ties in cumulative tournament points are shared equally.
- A player with no legal move: Not applicable — the Guard must select one valid suspect from the unrevealed set.
- Repetition / cycles: Bounded strictly by the single-accusation decision per round and the fixed round count .
- Simultaneous claims on the same resource or position: Not applicable — phases are strictly sequential.
- Player elimination mid-round; turn order repair: If an agent disconnects before role reveal, the round is aborted and re-dealt.
- Resource exhaustion (pieces, tokens, space, time): Role cards are finite ( cards); game ends after rounds.
- Termination guarantee: why the game provably ends: Guaranteed by the finite fixed round count and single-step accusation resolution.
## State representation
- Persistent state: Match round counter , global scoreboard `Dict[AgentID, int]`.
- Per-player state: `private_role: {JUDGE, GUARD, THIEF, CITIZEN}`, `is_revealed: bool`, `bluffing_skill: float`.
- Per-turn ephemeral state: Active phase `{DEALING, REVEAL, INTERROGATION, ACCUSATION, RESOLUTION}`, emitted signal vector , accused agent ID, round outcome classification.
- Observable information: **Partially observable** — Judge and Guard roles are public; unrevealed agents' roles are private to themselves.
## Action space
- `RevealRole()`: Publicly declare Judge or Guard identity. Precondition: Player holds the corresponding role.
- `Interrogate(suspect_id)`: Guard requests a behavioral signal probe from a suspect.
- `EmitSignal(bluff_effort)`: [gap-fill] Suspect chooses composure level to modulate emitted signal variance.
- `Accuse(suspect_id)`: Guard submits final indictment. Precondition: Suspect is unrevealed.
## Playability notes
- Single-player-vs-AI: Natural — 1 human player plays the Guard deducing the AI Thief from noisy text/metric signals, or plays the Thief attempting to blend in with AI Citizens.
- Group games (n>4): Standard format scales naturally from 4 to 12 agents; increasing expands the citizen suspect pool and raises deduction difficulty.
- Physical-skill games: Psychological bluffing, micro-expressions, and vocal tone are modeled as a noisy signal detection problem under Bayesian inference.
- What is lost in the adaptation: The spontaneous laughter, lively fireside banter (*degişme*), and comedic performance of folk penalty tasks are replaced by point tallies.
## Implementation hints
Low complexity. Bayesian probability model or discrete classification problem. Branching factor is choices for the Guard. Suggested first milestone: implement a 4-player game with Gaussian signal emission and a Bayesian Guard agent, testing true positive and false positive sentencing outcomes across 10 rounds.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
The Thief and the Judge (Turkmen Role-Play Deduction Game) (Turkmenistan). (n.d.). Games of Countries. https://gamesofcountries.com/games/tm-turkmenistan/ogry-kazy
@misc{gamesofcountries-tm-turkmenistan-ogry-kazy,
title = {The Thief and the Judge (Turkmen Role-Play Deduction Game) (Turkmenistan)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/tm-turkmenistan/ogry-kazy}
}