Formal Specification
The Shepherd's Ring — Formal Specification
Jersey · guessing-hidden · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: directGap-fills: 3
# La Bague bergère — AI-optimized specification
## General idea
A hidden-information deduction and bluffing circle game. A single token (the ring) moves sequentially along a closed ring graph among circle agents. Circle agents simultaneously execute noisy passing or bluffing actions each beat to obscure the token's position, while a central Seeker agent observes noisy movement signals and attempts to identify the token's exact location within a bounded number of guesses.
## Formal model
- **Abstraction taken:** The continuous song and physical hand-sliding along the string is abstracted into discrete synchronous passing rounds (ticks). Each tick, the token holder may either hold the token or pass it to an adjacent neighbor in the circle with a noisy bluffing signal generated by all agents.
- **Player/agent count:** agents (Agent 0: `SEEKER`, Agents : `CIRCLE_PLAYERS` arranged in cyclic order ).
- **Turn model:** Two-phase loop per beat:
1. Circle Phase: Simultaneous secret transfer/hold actions by circle agents.
2. Seeker Phase: Seeker inspects public noisy observation vector and either issues a formal accusation `accuse(target_id)` or calls `pass_beat()`.
## Formal rules
R1. Setup: circle agents are arranged in a ring graph with vertices , where neighbors of node are and .
- A single token `RING` is initially assigned uniformly at random to one circle agent: .
- Seeker has an allowance of incorrect guesses per round.
R2. Circle Action (Simultaneous):
- The agent secretly selects an action: .
- If , .
- If , .
- All other circle agents select dummy bluffing actions to mimic physical rhythm.
R3. Information Emission:
- At each tick , each circle agent emits an observable activity signal , where if agent was involved in an actual transfer, and otherwise, providing a noisy observation vector to the Seeker.
R4. Seeker Action:
- Seeker updates their Bayesian belief distribution .
- Seeker chooses either:
- `pass_beat()`: Advance to next passing tick .
- `accuse(target_id)`: Challenge circle agent .
R5. Accusation Resolution:
- If : Accusation is `SUCCESS`. Seeker scores point; the accused agent becomes the new `SEEKER`, and the previous Seeker joins the circle. Round terminates.
- If : Accusation is `MISS`. Guess allowance decreases: .
- If , play continues to tick .
- If , Seeker has exhausted guesses (`SEEKER_FAILED`). The circle team scores point, and a new round begins with the same Seeker [gap-fill: Seeker remains in center if failing 3 guesses].
R6. [gap-fill] Game Horizon: A full match consists of rounds. The winner is the agent who scored the highest individual points when acting as Seeker.
## Edge cases
- Tie at the end condition: If multiple agents have equal points at , the agent who found the ring in the fewest total ticks breaks the tie.
- A player with no legal move: Not applicable; circle agents always have legal transfer/hold actions, and the Seeker always has `pass_beat` or `accuse`.
- Repetition / cycles (and the loop-breaking rule): [gap-fill] A maximum of ticks is allowed per round. If the Seeker makes no accusation within 30 ticks, the round ends as an automatic `SEEKER_FAILED`.
- Simultaneous claims on the same resource or position: Not applicable; there is exactly one ring token and one Seeker accusation per beat.
- Player elimination mid-round; turn order repair: No player elimination; roles rotate upon successful accusation.
- Resource exhaustion (pieces, tokens, space, time): Only 1 ring exists; guess allowance is bounded (); ticks are bounded ().
- Termination guarantee: Guaranteed by the hard tick horizon per round and fixed round count .
## State representation
- Persistent state:
- `round`: Integer .
- `seeker_id`: Agent index .
- `scores`: Array of size .
- Ephemeral state (per tick):
- `tick`: Integer .
- `ring_holder`: Integer .
- `guesses_left`: Integer .
- `history_signals`: Matrix of past emitted signal vectors .
- Observable information:
- Circle agents observe their own position, whether they hold the ring, and who passed to them.
- Seeker observes only the public signal history , current tick, guesses remaining, and player IDs. The true `ring_holder` is hidden from the Seeker.
## Action space
- Circle Agent :
- `transfer(direction)`: Enum (`HOLD`, `LEFT`, `RIGHT`). (Only effective if ).
- Seeker:
- `pass_beat()`: Wait for next signal emission.
- `accuse(agent_id)`: Select accused circle agent index.
## Playability notes
- Single-player-vs-AI: Natural and compelling in both roles:
1. Human as Seeker: Solves a real-time signal filtering and POMDP challenge against AI circle bluffers.
2. Human as Circle Player: Manages timing of ring passes when Seeker's attention is focused elsewhere.
- Group games (n>4): The model naturally scales to any .
- Physical-skill games: Physical sleight of hand is parameterized by the signal-to-noise ratio: .
- What is lost in the adaptation: Vocal intonation, peripheral vision tracking, micro-expressions, and physical string tension changes.
## Implementation hints
- Implement as a discrete Hidden Markov Model (HMM) where the hidden state is the ring position and transition matrix is tridiagonal on the ring graph.
- Seeker AI can execute exact forward belief updating: .
- First milestone: Verify HMM belief tracking in a 4-agent circle before introducing dynamic seeker accusation thresholds.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
The Shepherd's Ring (Jersey). (n.d.). Games of Countries. https://gamesofcountries.com/games/je-jersey/la-bague-bergere
@misc{gamesofcountries-je-jersey-la-bague-bergere,
title = {The Shepherd's Ring (Jersey)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/je-jersey/la-bague-bergere}
}