Skip to main content
Formal Specification

Blind Man's Buff (Szembekötősdi / Kendőzés) — Formal Specification

Hungary · guessing-hidden · Confidence: high

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: discretizedGap-fills: 1
# Szembekötősdi — AI-optimized specification ## General idea A partial-observability pursuit and Bayesian identity-inference game where a sensory-deprived seeker navigates a bounded 2D arena using intermittent acoustic bearings, captures an elusive target agent, and executes an inference query over tactile/vocal feature vectors to correctly unmask the captive. ## Formal model - **Abstraction taken:** Visual observation for the Seeker is zero (`FOV = None`). The 2D circular arena (radius = 5m) is discretized into coordinates. Acoustic clues emitted by elusive agents provide noisy angle/distance vectors `(theta + epsilon, d + delta)`. Captured agent identity is represented by a feature vector `[height, hair_length, voice_pitch, clothing_texture]`. - **Player/agent count:** 1 Seeker `S` vs `N` Evaders `E_1..E_N` (standard `N = 4`). - **Turn model:** Tick-based simultaneous movement (tick = 0.25s) transitioning to a discrete Inference Stage upon capture. ## Formal rules R1. Setup: Arena bounded by circular boundary `r <= 5.0m`. Seeker `S` starts at `(0,0)` with blindfold active. Evaders `E_i` start at distinct positions `r in [2.0m..4.5m]`. R2. Movement & Acoustic Emission Loop: - At each tick, Evaders choose `move(dx, dy)` and optionally `emit_clue(intensity)`. - Boundary Violation: If an evader crosses `r > 5.0m`, evader is forced to freeze at boundary for 20 ticks. - Seeker receives noisy acoustic observation `z_t = [(angle_i + noise, dist_i + noise)]` for emitting evaders. - Seeker chooses `step(dx, dy)` and `reach_attempt()`. R3. Capture Condition: - If Seeker executes `reach_attempt()` while `dist(S, E_k) <= 0.8m`, Evader `E_k` is captured. Movement tick loop pauses. R4. Identification Phase (Inference Query): - Seeker receives noisy feature observations from captive `E_k`: - Tactile sample: `Obs_tactile ~ Normal(E_k.tactile_features, sigma_tactile)`. - Voice sample: Seeker issues command `voice_test()`; captive emits `Obs_voice ~ Normal(E_k.voice_pitch, sigma_voice * disguise_effort)`. - Seeker outputs identity claim `guess_id in [1..N]`. R5. Resolution: - If `guess_id == E_k.true_id`: Seeker succeeds. `E_k` becomes the new Seeker for the next round. Round terminates. - If `guess_id != E_k.true_id`: Seeker fails. Captive `E_k` is teleported/released to `r = 4.0m`. Movement loop resumes with same Seeker. R6. [gap-fill] Search Cap: If Seeker fails 5 consecutive identification trials, current round terminates with a penalty and next player in queue takes Seeker role. ## Edge cases - Tie at the end condition: Not applicable — round ends instantly on a single successful identification. - A player with no legal move: Seeker and evaders always have legal movement vectors within the bounded circle. - Repetition / cycles: Bounded by R6 penalty cap (max 5 failed trials). - Simultaneous claims on the same resource or position: Seeker can only grasp 1 evader per reach action (closest agent within reach radius). - Player elimination mid-round: When capture occurs, other evaders freeze during the identification trial; game resets or resumes per R5. - Resource exhaustion (pieces, tokens, space, time): Continuous coordinate arena; static feature distributions. - Termination guarantee: Enforced by the finite failure cap in R6 and non-zero posterior probability of correct identification. ## State representation - Persistent state: `current_seeker_id`, `seeker_penalties[agent_id]`, `round_number`. - Per-player static profile: `true_id`, `tactile_features [3]`, `voice_pitch`, `evasion_speed`. - Per-tick ephemeral state: `seeker_pos (x,y)`, `evader_positions[N] (x,y)`, `active_captive_id (or None)`, `failed_guesses_count`. - Observable Information: - Seeker: Zero visual state; receives only noisy acoustic vectors and post-capture feature samples. - Evaders: Fully observe Seeker position, own position, and boundary. ## Action space - Seeker: `move(dx, dy)`, `reach_out()`, `request_voice_sample()`, `submit_identity_guess(agent_id)`. - Evaders: `move(dx, dy)`, `emit_clap()`, `emit_disguised_voice(disguise_effort)`. ## Playability notes - Single-player-vs-AI: Highly playable as an audio-based stealth puzzle or probabilistic deduction game. - Group games (n>4): Excellent multi-agent benchmark for acoustic SLAM and particle-filter tracking. - Physical-skill games: Physical groping and vocal mimicry converted to noisy sensor vectors and Bayesian classifier updates. - What is lost: The visceral vulnerability of true physical sensory deprivation and tactile skin warmth. ## Implementation hints Low-to-medium complexity. Model the Seeker's brain as a Particle Filter tracking evader coordinates, coupled with a Gaussian Naive Bayes or softmax classifier for captive identity assignment. Suggested first milestone: implement the 2D movement and capture trigger, then attach the feature-sampling inference check.

Download as JSON

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

Download JSON
Cite this page

Blind Man's Buff (Szembekötősdi / Kendőzés) (Hungary). (n.d.). Games of Countries. https://gamesofcountries.com/games/hu-hungary/szembekotosdi

@misc{gamesofcountries-hu-hungary-szembekotosdi,
  title = {Blind Man's Buff (Szembekötősdi / Kendőzés) (Hungary)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/hu-hungary/szembekotosdi}
}