Formal Specification
Armenian hide-and-seek — Formal Specification
Armenia · hide-seek · Confidence: high
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: discretizedGap-fills: 2
# Pakhkvotsi — AI-optimized specification
## General idea
A base-tag hide-and-seek distinct for two structural reasons: spotting a hider does not itself catch them — it triggers a foot race back to base between the Seeker and the named hider — and a single dramatic exception, the "hero savior" rule, lets the last still-free hider undo every prior capture at once if they reach base first. Role rotation is a reward reserved specifically for a totally clean sweep, not for merely winning most exchanges.
## Formal model
- **Abstraction taken:** Real stealth, sprint speed, and visual search are abstracted into stochastic checks — a spotting probability while the Seeker searches, and a race-outcome probability once a hider is named — since the entry describes these qualitatively without numeric models. The play area is treated as bounded and finite, since the entry describes a village yard/garden setting, not an unbounded space.
- **Player/agent count:** 1 Seeker, K Hiders, K in [2, 14].
- **Turn model:** Synchronous ticks once the search phase begins; hiding and counting are sequential setup phases before it.
## Formal rules
R1. Seeker selection (entry §1.1): the initial Seeker is chosen by a counting-out rhyme, modeled as uniform random choice among all K+1 players (flavor rhyme has no further mechanical effect).
R2. Count phase (entry §2.1-2.2): the Seeker performs `count_to(T)` with face at the base, where T ∈ {30, 50, 100} is fixed once at session start [gap-fill: the entry offers three values without specifying which applies by default; treated as a session parameter, not a single invented number]. Hiders choose hiding positions during this phase, unobserved by the Seeker.
R3. Search phase: each tick, the Seeker submits `move(direction)` while hidden hiders remain unresolved. At each tick, resolve a stochastic `spot_check` for any hider within the Seeker's current search range; success sets that hider's status to `spotted`.
R4. Base race (entry §3.2-3.3): the instant a hider is `spotted`, both that hider and the Seeker race to the base — the hider via `sneak_to_base()` (if not already en route) and the Seeker via `sprint_to_base()` — resolving a stochastic race-outcome check. If the Seeker touches base first and correctly names the hider, that hider is `caught` and sits out near the base. If the hider reaches base first, they are `safe` for the round via ordinary self-rescue (entry §4.1, "Stuk indz!").
R5. Opportunistic self-rescue (entry §4.1): at any tick, any hider who is not yet `spotted`, `caught`, or `safe` may submit `sneak_to_base()` unprompted; if they reach base before the Seeker is there to block them, they become `safe`, independent of R3/R4.
R6. Hero savior (entry §4.2): if the hider who becomes `safe` under R4 or R5 is, at that instant, the *last* hider not yet `caught` or `safe`, all currently `caught` hiders are immediately freed (status reset to `safe`) and the round ends.
R7. Clean-sweep round end (entry §5.1): if every hider ends the round `caught` (no hider ever became `safe`, R6 never triggered), the round ends and the first hider who was caught becomes the new Seeker for the next round.
R8. Any-rescue round end (entry §5.2 and beyond): if the round ends with at least one hider `safe` but R6's hero condition was not met by that specific rescue (i.e., some hiders are `caught` and some `safe`, without a full team liberation), the same Seeker counts again for the next round [gap-fill: the entry explicitly defines only the two extreme outcomes — a totally clean sweep (§5.1) and a hero liberation (§5.2) — and is silent on a mixed outcome where some hiders are caught and others individually self-rescue without triggering R6; keeping the same Seeker is the conservative reading, since role rotation is explicitly framed as a reward for catching *everyone* cleanly].
## Edge cases
- Tie at the end: not applicable — R4's race check resolves to a single winner per race; there is no mechanism for a genuinely simultaneous base-touch in the entry's description.
- A player with no legal move: not applicable — the Seeker always has `move`/`sprint_to_base` available; any unresolved hider always has `sneak_to_base()` or continuing to hide available.
- Repetition / cycles: not applicable in the strict sense — each hider's status (`hidden` → `spotted`/`caught`/`safe`) only ever advances forward, never regresses, so no hider's state can cycle.
- Simultaneous claims on one position: not applicable — the base is a single point; R4's race resolves who arrives "first," and only one outcome (Seeker-wins or hider-wins) applies per race.
- Elimination mid-round and turn-order repair: a `caught` hider is removed from the active pool but can still be un-caught in bulk by R6; there is no turn order to repair since the search phase is simultaneous, not sequential.
- Resource exhaustion: not applicable — no consumable resource; hiding spots are reusable and not tracked as a depletable quantity.
- Termination guarantee: the play area is finite and bounded (R note in Formal model), and the Seeker's search is modeled as eventually exhaustive over it, so the `spot_check` (R3) probability for any remaining hidden hider does not vanish as search continues — every hidden hider is spotted with probability 1 given enough ticks. Once spotted, R4 resolves in a single race with a fixed positive probability of either outcome, and R5 gives hiders an independent path to resolution at any time. Since K is finite and every hider's status monotonically advances toward a terminal value (`caught` or `safe`) with non-vanishing per-tick probability, the round terminates with probability 1.
## State representation
- Persistent (session) state: `seeker_id`, `round_number`.
- Per-round state: `status[hider]` ∈ {hidden, spotted, caught, safe}, `caught_order` (list, to determine R7's "first caught").
- Per-tick ephemeral state: `positions` (Seeker and any hider currently racing), `search_coverage` (Seeker's cumulative explored area, informing `spot_check` probability).
- Observable information: the Seeker has no information about `hidden` hiders' positions until a `spot_check` succeeds; hiders have full information about the Seeker's position at all times (they are watching for an opportunity to rescue).
## Action space
- `count_to(T)`: the Seeker only, setup phase.
- `move(direction)`: the Seeker only, search phase; precondition: hiders remain `hidden` or `spotted`.
- `sprint_to_base()`: the Seeker only; precondition: a hider is `spotted` and unresolved.
- `sneak_to_base()`: any hider with status `hidden` or `spotted`; always available.
## Playability notes
- Single-player-vs-AI: natural on either side — human Seeker against AI hiders managing when to risk a self-rescue sprint, or human hider timing a dash against an AI Seeker's search pattern.
- Reducing K > 4 to 2 agents: mechanically trivial (fewer hiders), but removes the hero-savior drama, which specifically depends on there being a "last" hider among several.
- Physical-skill games: real stealth positioning, sprint speed, and visual search are the entire game and are collapsed into the `spot_check` and race-outcome stochastic checks (R3-R4); no genuine spatial reasoning or athleticism is modeled.
- What is lost: the actual tension of a stealth approach to base, the specific thrill of the last-hider hero sprint, the counting rhyme and ritual call-and-response phrases, and the social dynamic of a caught player waiting near base for potential liberation.
## Implementation hints
State is modest (a status enum per hider plus positions); the real design difficulty is calibrating `spot_check` and race-outcome probabilities as functions of position/concealment quality, neither of which the source quantifies. First milestone: implement the search-and-race loop (R3-R5) with flat probabilities to verify the clean-sweep (R7) and hero-savior (R6) round-end conditions, then add position-dependent spotting probability.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Armenian hide-and-seek (Armenia). (n.d.). Games of Countries. https://gamesofcountries.com/games/am-armenia/pakhkvotsi
@misc{gamesofcountries-am-armenia-pakhkvotsi,
title = {Armenian hide-and-seek (Armenia)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/am-armenia/pakhkvotsi}
}