Skip to main content
Formal Specification

Cantonese Blindfold & Tail Tag (Catch the Snake / Blind Man's Buff) — Formal Specification

Hong Kong · hide-seek · Confidence: low

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: discretizedGap-fills: 4
# Chuk Se Cung — AI-optimized specification ## General idea A spatial imperfect-information pursuit-evasion game simulating Cantonese blindfolded tag. One blindfolded agent (the Catcher) relies entirely on noisy auditory cues emitted by evaders to navigate a bounded grid arena, tag moving targets, and deduce their hidden identities, while evaders balance taunting noise requirements with evasion stealth. ## Formal model - **Abstraction taken:** Continuous physical space is mapped to a discrete 2D grid (10×10 cells). Physical blindfolding is formalized as an asymmetric visibility mask: the Catcher has zero visual line-of-sight and receives only noisy distance-attenuated sound vectors, while evaders observe the full grid. Physical tactile identification is modeled as an identity-guessing action upon co-location. - **Player/agent count:** 4 agents: 1 Catcher (C) and 3 Evaders (E1,E2,E3). - **Turn model:** Simultaneous-move discrete time ticks (t=1,2,…,Tmax⁡). In each tick, all agents submit directional displacement actions simultaneously. ## Formal rules R1. **Arena & Setup:** The arena is a discrete 10×10 grid bounded by coordinates (x,y)∈[0,9]×[0,9]. Catcher starts at center (4,4) or (5,5); evaders spawn at randomized distinct perimeter cells (x,y) such that distance(Ei,C)≥4. R2. **Noise Obligation:** Every evader must emit an audio ping at least once every 5 ticks (`PING` action) or whenever the Catcher invokes a mandatory callout action (`CALL_OUT`). R3. **Simultaneous Movement:** Each tick, every agent chooses an action from {STAY,NORTH,SOUTH,EAST,WEST}. Catcher speed is 1 cell/tick. Evader speed is 1 cell/tick. R4. **Tag Condition:** A tag occurs when the Catcher moves into the same cell as an Evader (pos(C)=pos(Ei)) or when their paths cross across adjacent edges in the same tick. R5. **Identification Protocol:** Upon a tag, the simulation pauses movement. The Catcher must submit an identity guess g∈{E1,E2,E3}. - If g=Etagged (correct guess): Catcher wins the round. The tagged Evader becomes the new Catcher for the next round. - If g≠Etagged (incorrect guess): The tagged Evader escapes, repelling to an unoccupied cell at Manhattan distance 2 from the Catcher, and gains 2 ticks of tag immunity. R6. **Boundary Penalty:** Any evader choosing a move that steps outside the [0,9]×[0,9] grid is disqualified and immediately counts as caught. R7. [gap-fill] Arena discretized to a uniform 2D grid (10×10) with orthogonal 4-way neighborhood connectivity. R8. [gap-fill] Auditory perception model: When an evader emits a sound event (ping or call response) at (xE,yE), the Catcher at (xC,yC) observes a noisy vector v→=(xE−xC+ϵx,yE−yC+ϵy) where ϵx,ϵy∼𝒩(0,σ2) with variance scaling with Euclidean distance. R9. [gap-fill] Identification limit: The Catcher is allowed exactly 1 identity guess per tag event; failure resets the search immediately. R10. [gap-fill] Loop-breaking round cap: If the Catcher fails to successfully capture and identify an evader within 100 ticks (Tmax⁡=100), the round terminates with a victory for the Evader team. ## Edge cases - **Tie at the end condition:** Not applicable; win condition is an atomic event triggered either by a successful identification (Catcher win) or expiration of the 100-tick cap (Evader team win). - **A player with no legal move:** All agents always have the legal move `STAY` available. - **Repetition / cycles:** Agents looping positions in the grid are strictly bounded by R10's 100-tick global clock. - **Simultaneous claims on the same resource or position:** If the Catcher occupies a cell containing multiple evaders simultaneously, the Catcher resolves an identity guess against a randomly sampled one of the co-located evaders. If multiple evaders enter the same non-catcher cell, co-location is permitted. - **Player elimination mid-round:** If an evader steps out of bounds, they are removed from the active set for the remainder of the round. Remaining evaders continue play without index re-ordering. - **Resource exhaustion:** Auditory pings are infinite; spatial grid is fixed; time ticks are bounded by Tmax⁡=100. - **Termination guarantee:** Guaranteed by the monotonic increment of the global tick counter t toward Tmax⁡=100, and absorbing round-end states upon correct identification. ## State representation - **Persistent state:** `grid_dimensions: (10, 10)`, `round_limit: 100`. - **Per-player state:** - `catcher_pos: (x, y)` - `evader_states: Dict[agent_id, {pos: (x, y), last_ping_tick: int, immunity_ticks: int, is_active: bool}]` - **Per-turn ephemeral state:** `current_tick: int`, `pending_actions: Dict[agent_id, Action]`, `audio_signals: List[{source_id: str, noisy_vector: (float, float), volume: float}]`. - **Observable information:** - Evaders observe complete world state: all agent coordinates and past audio signals. - Catcher observes only own position `catcher_pos`, boundary collision flags, audio signals received this tick, and game timer. Evader coordinates and true identities are strictly hidden. ## Action space - `move(direction)`: Parameters: `direction in {STAY, NORTH, SOUTH, EAST, WEST}`. Preconditions: target cell within [0,9]×[0,9]. - `ping()`: Parameters: none. Preconditions: agent is an active Evader. Emits an audio signal receivable by Catcher. - `call_out()`: Parameters: none. Preconditions: agent is Catcher; forces all active Evaders to respond with a ping on the subsequent tick. - `guess_identity(target_id)`: Parameters: `target_id in {E1, E2, E3}`. Preconditions: Catcher currently shares a cell with an Evader without immunity. ## Playability notes - **Single-player-vs-AI:** Highly playable. Human can play as the blindfolded Catcher (navigating via directional sound cues and radar pings) against heuristic/reinforcement-learning evaders, or as an Evader dodging an algorithmic tracker. - **Group games (n>4):** Scales cleanly from 4 agents to N agents by adding evader instances with distinct sound signatures. - **Physical-skill games:** Physical agility and tactile sensing are abstracted into probabilistic audio noise distributions (σ) and discrete grid movements. - **What is lost in the adaptation:** Tactile touch exploration, physical grappling, personal voice disguise nuances, and room acoustics are abstracted into mathematical noise vectors. ## Implementation hints - Represent the arena as a 2D coordinate space. - Implement simultaneous action submission using a two-phase update: collect actions -> resolve movement collisions -> evaluate tag & guess -> emit audio signals. - Branching factor: 5 movement choices per evader, 5 choices + callout for catcher (54=625 joint actions per tick). - First milestone: Headless 2-agent pursuit on a 10×10 grid where Catcher uses Kalman-filtered audio observations to close distance to a single evader.

Download as JSON

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

Download JSON
Cite this page

Cantonese Blindfold & Tail Tag (Catch the Snake / Blind Man's Buff) (Hong Kong). (n.d.). Games of Countries. https://gamesofcountries.com/games/hk-hong-kong/chuk-se-cung

@misc{gamesofcountries-hk-hong-kong-chuk-se-cung,
  title = {Cantonese Blindfold & Tail Tag (Catch the Snake / Blind Man's Buff) (Hong Kong)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/hk-hong-kong/chuk-se-cung}
}