Formal Specification
Blind Man's Buff (Blind Granny) — Formal Specification
Türkiye · hide-seek · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: discretizedGap-fills: 3
# Körebe — AI-optimized specification
## General idea
A blindfold hide-and-seek game modeled as an auditory-tracking and hidden-profile classification game. The blindfolded tagger (*ebe*) moves around a 2D bounded arena, tracking runners based on noisy sound signals. Once a runner is grabbed, the tagger must execute a classification query based on tactile and vocal clues to correctly identify the runner.
## Formal model
- **Abstraction taken:** The bounded room is represented as a 10x10 grid with walls at `x, y ∈ [-5.0, 5.0]`. [gap-fill] The blindfolded status is represented as a partial-information sensor state: the ebe can only observe noisy coordinate vectors of runners who emit sounds.
- **Player/agent count:** 1 active ebe `E`, and `K = N - 1` runners (where `K ∈ [3..12]`).
- **Turn model:** Tick-based simultaneous movement.
## Formal rules
R1. Arena and Positions:
- Bounded area: `x, y ∈ [-5.0, 5.0]`. Any runner stepping outside this area is immediately captured and becomes the new `Ebe`.
- `pos_E: Coordinate(x, y)` (position of Ebe).
- `pos_r: Array[K] of Coordinate(x, y)` (positions of runners).
R2. The Movement and Signal Phase:
- On each tick, all agents simultaneously declare a movement vector up to their maximum speed.
- [gap-fill] Each runner `r` has a 50% chance of emitting a sound signal (clap or shout) on a tick.
- The Ebe's information set `I_E` contains:
- A set of noisy signals: `S_r = (x_r + noise_x, y_r + noise_y)` for each runner who emitted a sound.
- `noise` is inversely proportional to `skill(E)`.
- The Ebe chooses a movement vector based on these signals.
R3. The Capture:
- If at any tick `distance(pos_E, pos_r) <= 0.5` meters for any runner `r`:
- Runner `r` is CAPTURED. All movements freeze.
- Game enters Phase 2 (Identification).
R4. The Identification Phase (The Guess):
- The Ebe executes a classification query.
- [gap-fill] The Ebe samples an identification feature set containing the runner's height, clothes texture, and voice pitch, with accuracy `P(Feature_Accuracy) = skill(E) * 0.70`.
- Based on this feature set, the Ebe declares a guessed name: `guess_id ∈ [0..K-1]`.
- Case 1: Correct Guess (`guess_id == r`).
- Runner `r` becomes the new `Ebe`.
- Round ends, positions reset, and a new round begins.
- Case 2: Incorrect Guess (`guess_id != r`).
- Runner `r` is released.
- Positions are offset by 2.0 meters (runners scatter).
- Phase 2 ends, the active play phase resumes.
R5. End condition:
- The game ends after a fixed time limit or after 5 successful role transitions.
R6. Win condition: The runner who avoided being the Ebe for the longest duration wins.
## Edge cases
- **Tie at the end condition:** Shared victory.
- **A player with no legal move:** Not applicable; agents can always choose a direction to crawl or run.
- **Repetition / cycles:** State is dynamic; if guesses fail, runners scatter, preventing static deadlocks.
- **Simultaneous claims on the same resource:** If the Ebe is within 0.5 meters of multiple runners, the runner with the minimum distance is captured.
- **Player elimination mid-round:** Not applicable.
- **Resource exhaustion:** Not applicable.
- **Termination guarantee:** Guaranteed by the fixed time limit in R5.
## State representation
- Persistent state:
- `pos_E: Coordinate(x, y)`.
- `pos_runners: Array[K] of Coordinate(x, y)`.
- `active_ebe_id: integer`.
- `game_phase: integer` (0: play, 1: identification).
- `captured_runner_id: integer`.
- Observable information:
- Runners observe all positions (perfect information).
- The Ebe observes only their own position and noisy signal coordinates.
## Action space
- `move_ebe(vector)`: declare movement vector. Valid only for Ebe in Phase 0.
- `move_runner(vector)`: declare movement vector. Valid only for runners in Phase 0.
- `guess_identity(runner_name)`: declare guessed name. Valid only for Ebe in Phase 1.
## Playability notes
- Single-player-vs-AI: Excellent. Can be played as the Ebe, where the player must guess which AI was caught based on a profile of AI characteristics, or as a runner dodging a blindfolded AI hunter.
- Group games: Generalizes perfectly to any number of players.
- Physical-skill games: Physical movement, noise-making, and blindfold searching are fully abstracted into coordinate grids, sensor noise, and classification queries.
- What is lost: The sensory deprivation of a real blindfold and the physical comedy of groping in the dark.
## Implementation hints
Medium complexity. Requires handling a partial-observation model for the Ebe (adding Gaussian noise to signal coordinates). First milestone: build the 2D boundary and movement updates, and implement the distance-based capture trigger.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Blind Man's Buff (Blind Granny) (Türkiye). (n.d.). Games of Countries. https://gamesofcountries.com/games/tr-turkiye/korebe
@misc{gamesofcountries-tr-turkiye-korebe,
title = {Blind Man's Buff (Blind Granny) (Türkiye)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/tr-turkiye/korebe}
}