Skip to main content
Formal Specification

Sheep and wolf — Formal Specification

Armenia · chase-tag · Confidence: medium

← Back to game page

Implementation Suitability

Suitability: adaptation-requiredFidelity: discretizedGap-fills: 3
# Vochkhar ev gayl — AI-optimized specification ## General idea A cooperative chain-defense chase distinct from an ordinary tag game in two ways: the Wolf may legally target only the rearmost sheep in an intact chain (or any sheep that has come loose), and the Shepherd's role is purely defensive positioning — mirroring the Wolf's lateral movement to physically block its path — rather than pursuit. The game has two independent, asymmetric win conditions: total capture for the Wolf, or a survival clock for the Shepherd's side. ## Formal model - **Abstraction taken:** Real flanking footwork, chain-swaying, and tag contact are abstracted into stochastic per-tick success checks whose probability is modulated by the Shepherd's blocking position, since the entry describes the dynamic qualitatively without a numeric model. - **Player/agent count:** 1 Wolf, 1 Shepherd, N Sheep, N in [4, 14], where Sheep are chained in a fixed line order behind the Shepherd. - **Turn model:** Synchronous ticks; all agents submit movement/blocking actions simultaneously each tick, resolved at tick end, bounded by a fixed match clock. ## Formal rules R1. Setup: the Shepherd stands at the head of the line; N Sheep form a single-file chain behind, each holding the waist/shoulders of the player ahead (entry §1.2); the Wolf starts 3-4 m facing the Shepherd (entry §1.3). `chain_intact = true` for all Sheep. R2. Opening dialogue (entry §2): flavor only, no mechanical effect on state. R3. Per-tick actions: the Wolf submits `move(direction)` (entry §3.1, feinting/darting laterally); the Shepherd submits `mirror_block(direction)`, attempting to match the Wolf's lateral position (entry §3.2); the Sheep chain submits a collective `sway(direction)`, swinging the tail of the line away from the Wolf (entry §3.3). R4. Chain integrity (entry §3.4): if the chain's swaying/movement fails to keep pace with the Wolf's positioning (a stochastic check per tick, since the entry gives no numeric threshold for "keeping the tail far enough away" [gap-fill]), one or more trailing Sheep become `detached` (`chain_intact = false` for them), making them individually vulnerable regardless of their position in the original line order. R5. Legal tag targets (entry §4.1): the Wolf's `tag()` action may only target either (a) the current rearmost Sheep in the still-intact portion of the chain, or (b) any `detached` Sheep. R6. Tag resolution: `tag()` against a legal target resolves a stochastic success check whose probability decreases the more closely the Shepherd's current `mirror_block` position matches the Wolf's position [gap-fill: the entry ties the Shepherd's blocking directly to preventing the Wolf's advance but gives no numeric relationship, so a position-alignment-sensitive probability is the minimal faithful reading]. R7. Capture (entry §4.2-4.3): a successful tag captures that Sheep, removing them from the chain and sending them to the Wolf's den; the chain immediately re-forms with the remaining Sheep, and the new rearmost (or any still-`detached`) Sheep becomes the next legal target (R5). R8. Wolf win (entry §5.1): the instant all N Sheep have been captured, the Wolf wins. The first Sheep captured becomes the new Wolf for the next round. R9. Shepherd win (entry §5.2): if a fixed duration elapses (entry's example, "2 minutes," is illustrative, not a fixed rule [gap-fill: treated as an external session parameter]) with zero Sheep captured, the Shepherd's side wins. ## Edge cases - Tie at the end: not applicable — R8 and R9 are mutually exclusive and jointly exhaustive: either the clock expires first (Shepherd win) or the Sheep count reaches zero first (Wolf win); no simultaneous-trigger case exists since capture count and elapsed time are independent, monotonic, non-reversing quantities. - A player with no legal move: not applicable — the Wolf always has `move`/`tag` (when a legal target exists) available; the Shepherd always has `mirror_block` available; Sheep always have `sway` available. - Repetition / cycles: not applicable — the match clock (R9) bounds total duration regardless of how many ticks pass without a capture; captures themselves are monotonically non-decreasing and bounded by N (R8). - Simultaneous claims on one position: not applicable — R5 restricts the Wolf to exactly one legal target category per tick (rearmost-intact or any-detached), so there is no ambiguity about which Sheep a successful tag applies to beyond the Wolf's own choice among currently-legal targets. - Elimination mid-round and turn-order repair: a captured Sheep is removed from the chain and the chain re-forms (R7) with no disruption to the remaining Sheep's relative order; there is no turn order to repair since all agents act simultaneously each tick. - Resource exhaustion: not applicable — no consumable resource; the chain and blocking mechanics are continuously reusable. - Termination guarantee: the match is bounded by two independent, always-progressing quantities: elapsed time (strictly increasing, capped at the fixed duration, R9) and captured-Sheep count (strictly non-decreasing, capped at N, R8). Whichever cap is reached first ends the match; since both are finite and monotonic, the match terminates within a fixed, bounded number of ticks regardless of the stochastic outcomes of individual tag attempts. ## State representation - Persistent (match) state: `captured_count`, `time_elapsed`, `wolf_id` (for the next round's reassignment, R8). - Per-tick ephemeral state: `chain_order` (list of still-in-chain Sheep, rearmost first), `detached_sheep` (set), `wolf_position`, `shepherd_position`. - Observable information: fully observable — all positions, chain state, and capture counts are public; no hidden information. ## Action space - `move(direction)`: the Wolf only; precondition: match ACTIVE. - `mirror_block(direction)`: the Shepherd only; precondition: match ACTIVE. - `sway(direction)`: the Sheep chain collectively; precondition: match ACTIVE. - `tag()`: the Wolf only; precondition: a legal target exists per R5. ## Playability notes - Single-player-vs-AI: natural on the Wolf side (human Wolf probing an AI Shepherd's blocking pattern) or the Shepherd side (human Shepherd defending against an AI Wolf); the Sheep chain's collective `sway` is best modeled as a single coordinated AI behavior rather than N independently-controlled agents. - Reducing N > 4 to 2 agents: reduces the chain to a minimal 2-sheep line, which removes the tactical tension of protecting a long vulnerable tail — a core part of the game's distinguishing structure. - Physical-skill games: real lateral footwork, chain-whipping coordination, and tag contact are the entire game and are collapsed into the position-sensitive stochastic tag check (R6); no genuine agility is modeled. - What is lost: the actual physical coordination of a swaying human chain, the ritual call-and-response dialogue, and the specific pastoral-survival stakes that gave the game its cultural resonance. ## Implementation hints State is modest (chain order, positions, a capture counter, a clock); the real design difficulty is calibrating how strongly the Shepherd's `mirror_block` alignment suppresses the Wolf's tag success probability (R6) and the chain-integrity check (R4), neither of which the source quantifies. First milestone: implement the tick loop with flat probabilities to validate the dual win conditions (R8-R9), then add position-sensitivity to blocking and chain integrity.

Download as JSON

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

Download JSON
Cite this page

Sheep and wolf (Armenia). (n.d.). Games of Countries. https://gamesofcountries.com/games/am-armenia/vochkhar-ev-gayl

@misc{gamesofcountries-am-armenia-vochkhar-ev-gayl,
  title = {Sheep and wolf (Armenia)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/am-armenia/vochkhar-ev-gayl}
}