Skip to main content
Formal Specification

The Watchman Knight — Formal Specification

Guernsey · singing-circle · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 2
# Le Chevalier du Guet — AI-optimized specification ## General idea A sequential team recruitment and line contest game. An initial single-agent Knight team engages an opposing line of N−1 Defenders across discrete verse rounds. Each round, the Knight team targets a specific Defender agent and initiates a 1v1 strength/persuasion contest; winning the contest transfers the target agent into the Knight's expanding line. The match terminates when all Defenders are absorbed. ## Formal model - **Abstraction taken:** The physical advancing lines and sung dialogue verses are abstracted into discrete sequential rounds of target selection followed by a stochastic contest resolver parameterized by agent strength/cohesion attributes. - **Player/agent count:** N total agents (N≥6). - **Turn model:** Strictly sequential rounds: Team A (Knight line) nominates a target → Contest resolves → Team memberships update. ## Formal rules R1. Setup: - N agents indexed {0,1,…,N−1}, each with attribute vector (strengthi,loyaltyi). - Initial team partitioning: Team K (Knight) ={0}, Team D (Defenders) ={1,2,…,N−1}. R2. Round Structure (Selection): - The leader of Team K selects a target agent j∈Team D. - The leader of Team D nominates a defending champion k∈Team D (often agent j themselves or the line captain). R3. Contest Resolution (Tug-of-War / Persuasion): - A stochastic contest is evaluated: P(capture)=∑i∈Kstrengthi+α∑i∈Kstrengthi+∑i∈Dstrengthi+2α, where α=1.0. - If capture succeeds: Target agent j is transferred from Team D to Team K: D←D∖{j}, K←K∪{j}. - If capture fails: Agent j remains in Team D; [gap-fill: Team K receives a +10% momentum bonus on the subsequent round to ensure non-stalling progression]. R4. Round Progression: - If |D|≥2, advance to next round R←R+1. R5. Terminal Condition and Scoring: - The game ends when |D|==1 (the final defender is absorbed by default). - Individual score for each agent i is calculated as the round index Rjoin,i when they joined Team K (the original Knight gets 0; the last remaining Defender gets N−1 points and becomes the new Knight). ## Edge cases - Tie at the end condition: The final remaining Defender uniquely earns the highest score; no ties for the primary winner. - A player with no legal move: Not applicable; as long as |D|≥2, Team K always has at least 2 valid targets in D. - Repetition / cycles (and the loop-breaking rule): [gap-fill] Due to the cumulative momentum bonus in R3, P(capture)→1 as failed attempts accumulate, guaranteeing transfer. Max round limit per game is capped at M=30. - Simultaneous claims on the same resource or position: Team K has a single unified target decision per round executed by the Knight line leader. - Player elimination mid-round; turn order repair: Agents are never eliminated; they transition team membership. - Resource exhaustion (pieces, tokens, space, time): The set of agents is constant; Team D strictly decreases in size toward the base case |D|=1. - Termination guarantee: Because each successful round permanently decreases |D| by 1, and P(capture) is strictly bounded away from 0, the state |D|=1 is an absorbing Markov state reached in finite time. ## State representation - Persistent state: - `round`: Integer counter ∈[1,30]. - `team_K`: Set of agent IDs. - `team_D`: Set of agent IDs. - `scores`: Map of agent ID → integer score. - `momentum`: Float value. - Ephemeral state (per round): - `target_id`: Chosen agent index j∈D. - `contest_result`: Boolean (`SUCCESS` / `FAILURE`). - Observable information: Fully observable public game state. ## Action space - Team K Leader: `select_target(agent_id)` where agent_id∈D. - Team D Leader: `select_defender(agent_id)` where agent_id∈D. ## Playability notes - Single-player-vs-AI: Playable as Defender (trying to survive as long as possible before capture) or as Knight (strategically targeting weaker defenders first to build Team K's total strength). - Group games (n>4): The multi-agent formulation captures social coalition and line-expansion dynamics naturally. - Physical-skill games: Physical tugging/singing is abstracted into attribute vectors and stochastic capture probabilities. - What is lost in the adaptation: Choral harmony, aesthetic dance choreography, and dramatic roleplay performance. ## Implementation hints - Very low state complexity: Two disjoint sets of integers representing team rosters. - Target selection policy for an AI Knight: Target arg⁡minj∈D⁡strengthj to maximize capture probability early, rapidly growing Team K's total strength advantage. - First milestone: Implement 6-agent simulation with random and greedy target selection policies, verifying monotonic shrinkage of Team D.

Download as JSON

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

Download JSON
Cite this page

The Watchman Knight (Guernsey). (n.d.). Games of Countries. https://gamesofcountries.com/games/gg-guernsey/le-chevalier-du-guet

@misc{gamesofcountries-gg-guernsey-le-chevalier-du-guet,
  title = {The Watchman Knight (Guernsey)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/gg-guernsey/le-chevalier-du-guet}
}