Formal Specification
Buffalo clash — Formal Specification
Armenia · strength-endurance · Confidence: high
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 5
# Gomshakriv — AI-optimized specification
## General idea
A locked-stance isometric pushing duel along a 1D line, distinct from a plain tug-of-war in that it explicitly permits two counter-tactics beyond raw force — sweeps (to unbalance the opponent regardless of position) and sidesteps (to redirect and blunt incoming force) — giving the contest a tactical layer on top of pure strength.
## Formal model
- **Abstraction taken:** Continuous isometric force and footing are abstracted into a 1D position on the line between the two rear boundaries, updated each tick by a force differential plus a stochastic slip term; ground-touch/collapse is a separate stochastic check triggered by sweep actions or accumulating fatigue, since the entry never gives a numeric force or balance model.
- **Player/agent count:** 2 agents per round.
- **Turn model:** Synchronous ticks; both agents submit an action simultaneously each tick, resolved at tick end.
## Formal rules
R1. Setup: both agents take a locked stance (hands-and-knees, crouched with shoulder grip, or forehead lock, entry §1.2) at `position = 0` on a line from `-2` (Agent A's rear boundary) to `+2` (Agent B's rear boundary), with the center line (`gits`) at 0 (entry §1.1, §1.3).
R2. On the referee's "Clash!" signal, the round enters ACTIVE phase.
R3. Each tick, both agents submit one action: `push()`, `sweep()`, `sidestep()`, or `hold()`.
R4. `push()` resolution (entry §2.1): position updates by a force differential (the pushing agent's committed force minus the opponent's) plus a stochastic slip term, moving `position` toward the opponent's rear boundary.
R5. `sweep()` (entry §2.2, "sweeps, leg trips... permitted to redirect momentum"): resolves a stochastic ground-touch check against the target agent, independent of the current `position` value [gap-fill: the entry permits sweeps as a redirection tactic but gives no numeric success model, so a direct stochastic check is the minimal formalization].
R6. `sidestep()` (entry §2.2, "evasive sidesteps... to redirect the opponent's momentum"): reduces the effective force differential applied against the sidestepping agent on that tick, acting as a defensive counter to an opponent's simultaneous `push()` [gap-fill: no numeric reduction factor is given].
R7. Illegal actions (entry §2.3): `strike`, `bite`, `gouge`, `throat_press`. Any of these committed by an agent is an immediate round loss for that agent, with no warning tier.
R8. Round loss conditions (entry §3.1-3.2): an agent loses the round the instant either (a) it touches the ground with head/face or fully collapses onto its belly (typically the outcome of a successful `sweep()` against it, R5, but can also occur from a large force imbalance under R4 [gap-fill: entry does not separate collapse-from-sweep and collapse-from-overwhelmed-push as distinct triggers, so both feed the same ground-touch state]), or (b) its `position` crosses its own rear boundary (`-2` for A, `+2` for B).
R9. Match structure (entry §4): best-of-three rounds; the match ends the instant one agent has won 2 rounds, who is declared "Village Buffalo." If both agents trigger a loss condition on the same tick, the round is a draw and replayed [gap-fill: entry is silent on simultaneous double-loss; least-invented resolution].
## Edge cases
- Tie at the end: handled at round level (R9's draw-and-replay clause); never a match-level tie since the match requires a clean 2-round majority.
- A player with no legal move: not applicable — `push`, `sweep`, `sidestep`, and `hold` are always available to an agent still in ACTIVE phase.
- Repetition / cycles: not structurally prevented by position alone (two evenly-matched agents holding could in principle keep `position` near 0 indefinitely), but `sweep()`'s ground-touch probability and accumulating fatigue (both increasing with elapsed ticks, mirroring the calibration duel's hazard model) drive the probability of an unresolved stalemate toward 0 as tick count grows [gap-fill: fatigue-scaling is assumed by analogy with the calibration game, since the entry describes the contest as strength/endurance-based but gives no explicit fatigue mechanic].
- Simultaneous claims on one position: not applicable — this is a 1D contest with a single shared position value, not a multi-cell occupancy game.
- Elimination mid-round and turn-order repair: not applicable — a round has exactly one loser and ends immediately (R8); the match's fixed best-of-three structure (R9) is unaffected by any single round's outcome beyond the running win count.
- Resource exhaustion: not applicable — no consumable resource; fatigue is a probability modifier, not a depletable quantity.
- Termination guarantee: `position` is confined to a bounded interval `[-2, 2]` and any sustained `push()` advantage drives it toward a boundary (R4, R8b); independently, the ground-touch hazard from `sweep()` and rising fatigue is non-decreasing over time and bounded away from 0 after any finite number of ticks, so the probability the round remains undecided after t ticks tends to 0 as t → ∞. The match is a fixed sequence of at most 3 such rounds (R9).
## State representation
- Persistent (match) state: `rounds_won[A]`, `rounds_won[B]`, `round_number`.
- Per-round state: `position` (shared, in `[-2, 2]`), `ticks_elapsed`, `ground_contact[agent]` (bool).
- Observable information: fully observable — both agents see the shared position and elapsed time; no hidden information.
## Action space
- `push()`: either agent; precondition: round ACTIVE, agent has not lost.
- `sweep()`: either agent; precondition: round ACTIVE.
- `sidestep()`: either agent; precondition: round ACTIVE.
- `hold()`: either agent; always legal while round ACTIVE.
- Illegal actions the referee model must recognize but that are never in the legal set: `strike`, `bite`, `gouge`, `throat_press` (R7).
## Playability notes
- Single-player-vs-AI: natural, one side of the duel vs an AI opponent managing its own push/sweep/sidestep mix.
- Physical-skill games: the entire isometric strength and balance contest is reduced to a 1D position value plus two stochastic checks (R5 ground-touch, R4 slip); no true force or footing physics is modeled.
- What is lost: the real neck/back/leg power differential between competitors, genuine tactical timing of a sweep or sidestep, and the livestock-fair festival framing with its praise songs for the winner.
## Implementation hints
State is minimal (one shared position value plus small per-round counters), so this is computationally trivial; the actual design difficulty is calibrating the push force differential, sweep ground-touch probability, and sidestep force-reduction factor so that sweeps and sidesteps feel like genuine tactical choices rather than dominant or irrelevant strategies. First milestone: implement `push()`-only rounds to validate the boundary-crossing loss condition (R8b) and best-of-three scoring (R9), then add `sweep()`/`sidestep()` and fatigue-scaling.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Buffalo clash (Armenia). (n.d.). Games of Countries. https://gamesofcountries.com/games/am-armenia/gomshakriv
@misc{gamesofcountries-am-armenia-gomshakriv,
title = {Buffalo clash (Armenia)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/am-armenia/gomshakriv}
}