Skip to main content
Formal Specification

Fish Trap — Formal Specification

Thailand · guessing-hidden · Confidence: low

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: discretizedGap-fills: 5
# Phong Phang — AI-optimized specification ## General idea A traditional Thai circle singing, blindfold tagging, and identity-deduction game based on estuary fish traps (*Phong Phang*). One blindfolded player represents a "blind fish" trapped inside a circle of net trappers; after a rhyming chant, the fish announces "Living Fish" (trappers can dodge in place) or "Dead Fish" (trappers must remain motionless); the fish then explores blindly, makes physical contact, and attempts to correctly identify the captured player through tactile trait clues. ## Formal model - **Abstraction taken:** The circular net enclosure is discretized into `N` positions around a central search grid. The blindfold condition is formalized as asymmetric hidden information: the Fish agent cannot observe player positions directly, receiving only spatial proximity signals. Tactile inspection generates a feature vector of clues (height, texture, frame) used to deduce the target player's identity from a known roster. - **Player/agent count:** 1 Fish agent vs `N` Trapper agents (`N in [4, 10]`). - **Turn model:** Phase 1: Simultaneous chant and mode declaration; Phase 2: Sequential search and deduction ticks by the Fish agent. ## Formal rules R1. Setup: `N` Trapper agents form a circle around a central 5x5 grid (`[-2, 2] x [-2, 2]`). Trappers are assigned fixed seats `0` to `N-1` on the outer rim. Each player has a public feature vector: `{ height_tier: [1..3], frame: [1..3], clothing_texture: [1..3] }`. R2. [gap-fill] Chant & Mode Declaration Phase: - Trappers complete the chant. - The Fish agent declares the mode: `Kin_Pla_Pen` (Living Fish) or `Kin_Pla_Tai` (Dead Fish). - If `Kin_Pla_Tai` is chosen: Trappers cannot change their coordinate or dodge; any movement is a foul making them instantly detectable. - If `Kin_Pla_Pen` is chosen: Trappers may execute a local dodge (`offset in {-1, 0, 1}`) while remaining seated. R3. [gap-fill] Blind Search Phase: - The Fish agent starts at origin `(0, 0)` with a budget of 15 movement ticks. - Each tick, Fish chooses a direction: `Move(dx, dy)` where `dx, dy in {-1, 0, 1}`. - If Fish enters a cell adjacent to a Trapper, contact is initiated (`contact = true`). R4. [gap-fill] Tactile Inspection & Deduction: - Upon contact, the Fish inspects the trapped player for 1 tick, observing 2 out of 3 features from the player's true feature vector (with 10% observation noise). - The Fish submits an identity guess: `Guess_Identity(player_id)`. - If `Guess_Identity` is correct: The identified player becomes the Fish for the next round. The guessing Fish scores 2 points. - If `Guess_Identity` is incorrect: The Fish scores 0 points, incurs 1 penalty, and remains the Fish for another round. R5. [gap-fill] Search Timeout: - If the Fish fails to make contact within 15 movement ticks, the search expires; the Fish scores 0 points and remains the Fish for the next round. R6. [gap-fill] Match Structure & End Condition: - The match consists of `R = 6` rounds. - The player with the highest score (or lowest penalties) at the end of 6 rounds is declared the overall winner. ## Edge cases - Tie at the end condition: Players with equal highest scores after 6 rounds share victory. - A player with no legal move: Not applicable; Fish always has movement/guess choices, and Trappers have freeze/dodge responses. - Repetition / cycles: Bounded by the 15-tick search limit per round and the 6-round hard match cap. - Simultaneous claims on the same resource or position: The Fish makes contact with the first Trapper encountered along their movement ray; if multiple Trappers are adjacent, the closest by Euclidean distance is selected for tactile inspection. - Player elimination mid-round; turn order repair: Non-elimination party game; role of Fish transfers dynamically upon a correct deduction without changing the player roster. - Resource exhaustion: Search movement budget (15 ticks) depletes strictly each turn; rounds count up to 6. - Termination guarantee: why the game provably ends: Each round has a deterministic upper bound of 16 ticks (15 search ticks + 1 deduction tick), guaranteeing the 6-round match terminates in at most 96 ticks. ## State representation - Persistent state: - `round_number`: integer in `[1, 6]` - `player_scores[N+1]`: integer array of points - `current_fish_id`: player ID - `player_features[N+1]`: dictionary mapping player IDs to feature vectors - Per-turn ephemeral state: - `fish_pos`: `(x, y)` coordinate in 5x5 grid - `trapper_positions[N]`: array of 2D coordinates - `game_mode`: `kin_pla_pen | kin_pla_tai` - `search_ticks_remaining`: integer in `[0, 15]` - `contact_target_id`: `null | player_id` - Observable information: - Trappers observe: full state (Fish position, mode, features). - Fish observes: only own position `(x, y)`, declared mode, audio cues (if foul in dead fish mode), and partial feature vector upon contact. The exact position and identity of surrounding players are hidden. ## Action space - Fish agent: - `declare_mode(mode)`: `mode in {kin_pla_pen, kin_pla_tai}`. - `move(dx, dy)`: Step in grid direction. - `inspect_and_guess(player_id)`: Inspect contacted target and submit guessed player ID. - Trapper agents: - `freeze()`: Remain motionless (mandatory in `kin_pla_tai`). - `dodge_lean(direction)`: Shift body within seated cell (allowed in `kin_pla_pen`). ## Playability notes - Single-player-vs-AI: Highly engaging Bayesian deduction challenge; AI Fish maintains belief distributions over player identities given observed feature clues. - Group games (n>4): Naturally supports 5 to 11 players in a circle. - Physical-skill games: Tactile texture sensing, blind navigation, and silent breath control are modeled via hidden information states and Bayesian feature sampling. - What is lost in the adaptation: The suspenseful singing of the *Phong Phang* fishing rhyme (*"Pla ta-bod khao lod phong phang..."*) and the humor of disguised voice squeaks. ## Implementation hints - Implement as a partially observable Markov decision process (POMDP) for the Fish agent. - Keep the feature space discrete and distinct to allow logical deduction even with partial information. - Suggested first milestone: Implement the 5x5 grid search with blind movement constraints and verify Bayesian identity deduction scoring across 6 rounds.

Download as JSON

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

Download JSON
Cite this page

Fish Trap (Thailand). (n.d.). Games of Countries. https://gamesofcountries.com/games/th-thailand/phong-phang

@misc{gamesofcountries-th-thailand-phong-phang,
  title = {Fish Trap (Thailand)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/th-thailand/phong-phang}
}