Formal Specification
Blind buck (Blindman's buff) — Formal Specification
Sweden · chase-tag · Confidence: high
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: discretizedGap-fills: 6
# Blindbock — AI-optimized specification
## General idea
An asymmetric imperfect-information tag game where a single sensory-deprived agent ("Blindbock") operates with zero visual observations in a bounded arena, navigating exclusively by acoustic noise cues emitted by fleeing sighted agents. Upon making physical contact with an evasion agent, the Blindbock performs a tactile feature-identification query; a correct identity prediction transfers the blindfold role to the captured agent.
## Formal model
- **Abstraction taken:** Continuous physical room space is discretized into an grid. Visual occlusion (the blindfold) is formalized as an empty visual observation set for the Blindbock, replaced by an acoustic emitter-receiver model where agent movement generates directional sound vectors with intensity inversely proportional to Euclidean distance. Tactile identification is modeled as sampling from a static per-agent feature vector subject to sensory noise.
- **Player/agent count:** agents (; default benchmark : 1 Blindbock , 5 Evasion agents ).
- **Turn model:** Synchronous discrete time ticks ().
## Formal rules
R1. **Setup:** The arena is an bounded grid. Each agent possesses an immutable public feature vector representing height, hair texture, and clothing. Agent is designated `BLINDBOCK` and placed at the center ; remaining agents are dispersed randomly within bounds.
R2. **Movement & Noise Emission:**
- On each tick , all agents simultaneously select an action: `move(dx, dy)` () or `stand_still()`.
- Moving generates a sound signature with amplitude ; standing still emits residual sound [gap-fill].
- Blindbock receives an aggregated acoustic signal .
R3. **Capture Contact:**
- If at the end of tick , for some evasion agent , is caught and enters state `HELD`.
- Both and are frozen for the identification phase.
R4. **Tactile Identification Query:**
- receives a noisy tactile measurement , where [gap-fill].
- outputs a prediction .
R5. **Identification Resolution:**
- If (correct guess): becomes the new `BLINDBOCK` for the next round. The round terminates with a win for .
- If (incorrect guess): is released with a 3-tick immunity window [gap-fill]. resumes searching.
- [gap-fill] If accumulates 3 consecutive incorrect guesses on caught players, a mandatory vocal ping is emitted by all evasion agents.
R6. **Round Timeout:**
- [gap-fill] If ticks elapse without a successful capture and identification, the round ends in a draw, and the role rotates to the next sequential agent ID.
## Edge cases
- Tie at the end condition: Not applicable — only one agent can be grabbed and identified per tactile phase.
- A player with no legal move: Sighted agents blocked by walls or other agents may choose `stand_still()` or reverse direction. Blindbock collisions with boundary walls result in zero displacement and a haptic wall-bump signal.
- Repetition / cycles: Bounded by the finite tick limit .
- Simultaneous claims on the same resource or position: Multiple evasion agents may occupy the same cell; if Blindbock enters a multi-occupant cell, one agent is selected uniformly at random for the capture query [gap-fill].
- Player elimination mid-round: If an agent disconnects, they are removed from the candidate set; if the active Blindbock disconnects, the lowest indexed evasion agent assumes the blindfold.
- Resource exhaustion (pieces, tokens, space, time): Spatial boundaries prevent leaving the arena. Time bounded by .
- Termination guarantee: Each round terminates upon a correct identification or reaching ticks.
## State representation
- **Persistent state:**
- `agent_features`: array of vectors .
- `current_blindbock`: agent ID.
- `round_number`: integer.
- `cumulative_scores`: map of agent ID .
- **Per-turn ephemeral state:**
- `tick`: integer .
- `positions`: map of agent ID .
- `held_agent`: optional agent ID.
- `incorrect_guess_count`: integer.
- **Observable information:**
- Evasion agents observe: full global positions, arena boundaries, and identity of all players.
- Blindbock observes: own position , boundary bump flags, aggregated acoustic vector , and tactile sample only when `held_agent` is set.
## Action space
- `move(dx, dy)`: submit vector .
- `guess_identity(target_id)`: submit predicted agent ID during tactile resolution phase.
## Playability notes
- Single-player-vs-AI: Highly compelling as an asymmetric audio-spatial simulation — human as Blindbock using 3D spatial audio cues to hunt AI agents, or human as an evader managing noise vs distance against a tracking AI hunter.
- Group games (n>4): Reduces to 2 agents (1 Blindbock vs 1 Evader), with the identification task replaced by a distance-based or posture-based classification challenge.
- Physical-skill games: Physical balance, tactile palpation, and auditory localization are formalized into vector feature classification and distance-decay signal processing.
- What is lost in the adaptation: The tactile warmth of touching real clothing/hair, the parlor laughter, and real-world spatial orientation challenges are replaced by numerical noisy estimation.
## Implementation hints
Moderate complexity. The acoustic propagation model requires a simple distance calculation per tick. Feature classification can use normalized attributes (e.g. ). First milestone: implement Blindbock navigation towards a single sound emitter in a grid using gradient descent on the acoustic vector.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Blind buck (Blindman's buff) (Sweden). (n.d.). Games of Countries. https://gamesofcountries.com/games/se-sweden/blindbock
@misc{gamesofcountries-se-sweden-blindbock,
title = {Blind buck (Blindman's buff) (Sweden)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/se-sweden/blindbock}
}