Skip to main content
Formal Specification

Blind man's buff — Formal Specification

Armenia · chase-tag · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 6
# Achk-kapuk — AI-optimized specification ## General idea A blindfolded-tag game whose distinctive mechanic is not capture but *identification*: touching an evader is easy, but the seeker must then name them correctly using touch alone while the captured player is actively disguising their posture and forbidden from speaking. There is no score and no defined endpoint — roles simply rotate forever whenever an identification succeeds, which makes this a perpetual social activity rather than a game with a terminal state. ## Formal model - **Abstraction taken:** The seeker's real sensory deprivation and tactile search is abstracted into two decoupled mechanics: ordinary tick-based movement/contact (for the chase) and a single stochastic "identify check" resolved once an evader is captured and frozen (for the guess). This is a probabilistic proxy, not a simulation of touch. - **Player/agent count:** 1 Seeker, N Evaders, N in [3, 11]. - **Turn model:** Synchronous ticks during the free-evasion phase (all agents move simultaneously); on capture, movement freezes for both the Seeker and the captured Evader and a single-shot identification resolution occurs before the next round's free-evasion phase resumes. ## Formal rules R1. Setup: one agent is selected Seeker uniformly at random among all N+1 players (entry §1.1's counting-out rhyme is a real-world randomization procedure, modeled here as uniform choice). The Seeker's state is set `blindfolded = true` for the remainder of the round; being spun three times before play (entry §1.3) has no mechanical effect on state [gap-fill: the spin/rhyme is flavor with no formalized disorientation penalty, since the entry never ties it to a search-ability parameter]. R2. Free evasion: each tick, every Evader submits `move(dx, dy)` within the bounded arena, or `wait()`, or `auditory_tease()` (clap/snap/call, entry §2.2). `auditory_tease()` has no listed direct effect on positions but may leak information to the Seeker [gap-fill: modeled as giving the Seeker a probabilistic bearing hint toward the signalling Evader on that tick, since the entry says the calls exist to "guide and tease the seeker" but never quantifies a detection radius]. Evaders may not move outside the arena boundary (entry §2.3); an attempted move past the boundary is clamped to the boundary edge rather than penalized [gap-fill: the entry states the boundary is "prohibited" to cross but specifies no penalty]. R3. Capture: the Seeker submits `grab()` when an Evader occupies an adjacent/contact cell. On a successful grab, both the Seeker and the grabbed Evader enter `frozen = true` and the identification phase begins; all other Evaders continue free evasion (R2) unaffected. R4. Identification: while frozen, the Seeker may repeat `tactile_examine()` any number of times (entry §3.2 lists head, hair, shoulders, clothing as examined features, with no limit given) and the captured Evader may repeat `disguise()` (hunch, alter posture, fluff hair) but may never take a `speak()` action (entry §3.3 forbids it). When the Seeker submits `declare_identity(name)`, resolve a stochastic identify-check: success probability `p_id` starts at a high baseline (touch-based identification with no disguise is easy) and decreases with the number of `disguise()` actions the captured Evader performed before the declaration [gap-fill: the entry confirms disguise lowers identifiability qualitatively — "attempted comical physical disguises" — but gives no numeric curve]. R5. Resolution (entry §4): if `declare_identity` is correct, the captured Evader becomes the new Seeker, the old Seeker rejoins the Evader pool, both `frozen` flags clear, and a new round's free-evasion phase (R2) begins immediately. If incorrect, the captured Evader is released (`frozen = false`, rejoins evasion) and the *same* Seeker continues searching; no role change occurs. R6. No score is tracked and no Evader is ever permanently removed from play (entry §5: "played in continuous rounds as a lighthearted... parlor game"). There is no win or loss condition for any agent. ## Edge cases - Tie at the end: not applicable — there is no scoring and no terminal state to tie at (R6). - A player with no legal move: not applicable — an unfrozen Evader always has `move`/`wait`/`auditory_tease`; the Seeker always has `move`/`wait` when unfrozen, or `tactile_examine`/`declare_identity` when frozen with a captive. - Repetition / cycles: possible and not structurally prevented. The formal model has no per-round tick limit, and the brief here deliberately avoids inventing one; a Seeker who never grabs anyone, or who grabs and consistently misidentifies (R5, incorrect branch), can in principle continue indefinitely. This is the honest consequence of R2-R5 as written from the entry, not an oversight. - Simultaneous claims on one position: if two Evaders are simultaneously in contact range on the tick the Seeker submits `grab()`, exactly one is captured; which one is chosen arbitrarily [gap-fill: the entry's capture rule (§3.1) only ever describes grabbing a single player, so a tie-break rule is invented to keep `grab()` well-defined]. - Elimination mid-round and turn-order repair: not applicable — no agent is ever removed from the game; "elimination" in this game is role reassignment (R5), not removal. - Resource exhaustion: not applicable — there is no consumable resource; the arena and player count are fixed and non-depleting. - Termination guarantee: **not guaranteed.** The source entry gives this game no win condition and describes it explicitly as continuous, role-rotating play (§5). Even a single round has no enforced end: R2-R5 as derived from the entry impose no tick limit on evasion and no cap on repeated misidentification, so indefinite play is a valid outcome of the formal model, not just a flavor note. `termination_guaranteed: false` reflects this honestly rather than inventing a stopping rule the entry does not support. ## State representation - Persistent state: `seeker_id` (current), informal `round_number` (does not gate any rule). - Per-round ephemeral state: `positions`: map of agent ID → (x, y); `frozen[agent]`: bool, true only for the Seeker and the currently captured Evader during identification; `disguise_actions_taken`: counter for the currently captured Evader, reset each time a new capture begins. - Observable information: the Seeker has **no positional information** of any agent (blindfolded) except whatever bearing hint an `auditory_tease()` leaks on the tick it is used [gap-fill: , see R2]. All Evaders have full positional information, including the Seeker's location and blindfolded status (which is public). ## Action space - `move(dx, dy)`: any unfrozen agent; precondition: destination within arena boundary (else clamped, R2). - `wait()`: any unfrozen agent; always legal. - `auditory_tease()`: Evaders only; precondition: unfrozen. - `grab()`: Seeker only; precondition: unfrozen, an Evader in contact range. - `tactile_examine()`: Seeker only; precondition: `frozen = true` with a captured Evader present; repeatable. - `disguise()`: the captured Evader only; precondition: `frozen = true`; repeatable. - `declare_identity(name)`: Seeker only; precondition: `frozen = true` and at least one `tactile_examine()` has been performed; resolves R4/R5 and ends the identification phase. ## Playability notes - Single-player-vs-AI: natural on the Evader side (human evades an AI Seeker). Awkward on the Seeker side for a digital adaptation, since real touch-based search has no direct screen analogue; it would need to be re-abstracted as a proximity/pattern-guessing minigame rather than literal "tactile examination." - Reducing N > 4 to 2 agents (1 Seeker, 1 Evader): mechanically trivial, but discards the multi-evader teasing dynamic and the disguise/decoy texture that gives the source game its social character. - Physical-skill games: the entire sensory search-and-identify skill is collapsed into the single stochastic parameter `p_id`; no genuine auditory localization or fine tactile discrimination is modeled. - What is lost: the real sensory deprivation, the physical comedy of disguising oneself while frozen and silent, the communal call-and-response teasing, and the fact that in reality the Seeker's search is driven by continuously-updating auditory cues rather than a single per-tick hint. ## Implementation hints Branching factor per tick is small and bounded (a handful of discrete actions per agent), so this is computationally trivial; the actual design work is entirely in choosing `p_id`'s disguise-sensitivity curve and whether/how `auditory_tease()` leaks bearing information, since the source entry specifies neither numerically. First milestone: implement free evasion + grab + a flat (non-disguise-sensitive) `p_id` with no auditory info leak, verify role-rotation on correct ID and continued search on incorrect ID (R5), then layer in disguise-sensitivity and tease-based hints once those parameters are chosen.

Download as JSON

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

Download JSON
Cite this page

Blind man's buff (Armenia). (n.d.). Games of Countries. https://gamesofcountries.com/games/am-armenia/achk-kapuk

@misc{gamesofcountries-am-armenia-achk-kapuk,
  title = {Blind man's buff (Armenia)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/am-armenia/achk-kapuk}
}