Skip to main content
Formal Specification

Prisoner's Base (Greek Base Tag) — Formal Specification

Greece · chase-tag · Confidence: high

← Back to game page

Implementation Suitability

Suitability: engine-readyFidelity: discretizedGap-fills: 2
# Ambariza — AI-optimized specification ## General idea A discrete, two-team spatial territory and tagging game played on a 2D grid between two fortified home bases. Tagging priority between opponents is strictly determined by departure timestamps: the agent that left their home base most recently possesses tagging dominance over an opponent who departed earlier. Teams balance base defense, distraction decoys, prisoner rescue chains, and direct base invasion. ## Formal model - **Abstraction taken:** Continuous sprinting and tagging field discretized to a 2D grid (W×H, e.g., 30×15 cells). Time advances in discrete simultaneous ticks. - **Player/agent count:** 2 teams (Team 0 and Team 1), with K agents per team (K∈[2..8], default K=4). - **Turn model:** Real-time ticked; all active agents submit directional movements simultaneously each tick. ## Formal rules R1. Board setup: Grid of size W×H. - Team 0 Base: Column x=0; Team 0 Prison: Cells (0,0..2). - Team 1 Base: Column x=W−1; Team 1 Prison: Cells (W−1,H−3..H−1). - Neutral Field: Columns 1≤x≤W−2. R2. Agent State: Each agent i has `(x, y)`, `status` ∈{in_base,active,jailed,safe_return}, and integer `exit_timestamp`. R3. Movement per tick: - Active agents can move 1 cell in any of 4 cardinal directions or stay in place: Δx,Δy∈{(−1,0),(1,0),(0,−1),(0,1),(0,0)}. - An agent touching their home base can set `status = in_base` and reset `exit_timestamp = current_tick`. R4. Tagging & Priority Resolution: - If active agent A (Team 0) and active agent B (Team 1) occupy the same cell or cross adjacent edges during a tick: - If `A.exit_timestamp > B.exit_timestamp`: Agent A tags B. Agent B transitions to `jailed` at Team 0 Prison. - If `B.exit_timestamp > A.exit_timestamp`: Agent B tags A. Agent A transitions to `jailed` at Team 1 Prison. - [gap-fill] If timestamps are identical, neither tags; both bounce back to previous tick positions. R5. Prison Chain & Rescue: - Jailed agents form a connected chain if each jailed agent is adjacent to another jailed agent, with at least one adjacent to the prison cell. - If an active friendly agent enters any cell occupied by a jailed teammate in the chain, all linked jailed teammates enter `safe_return` status (impervious to tags while walking directly to their home base). R6. Win Conditions: - Base Touch Win: An active agent lands on any cell of the opponent's home base → Immediate win for that agent's team. - Elimination Win: All K agents of one team are simultaneously `jailed` → Immediate win for the capturing team. R7. [gap-fill] Turn Limit: If neither win condition is satisfied within 500 ticks, the team with fewer jailed agents wins; if tied, match is a draw. ## Edge cases - Tie at the end condition: Resolved by R7 (point count of free agents or formal draw at tick 500). - A player with no legal move: Not applicable — waiting/holding position is always a legal action. - Repetition / cycles: Bounded by the hard 500-tick game clock (R7). - Simultaneous claims on the same resource or position: Friendly agents can co-occupy cells. Hostile encounters are strictly resolved via the deterministic `exit_timestamp` priority (R4). - Player elimination mid-round; turn order repair: Jailed agents skip active movement; remaining agents continue on simultaneous ticks without turn order disruption. - Resource exhaustion: Not applicable — no consumable resources. - Termination guarantee: Enforced by the finite tick horizon (T≤500) and invariant state transitions. ## State representation - Persistent state: - `agents`: Array of 2K structs: `[team, x, y, status, exit_tick]`. - `current_tick`: Integer 0..500. - `grid_dims`: (W,H). - Per-tick ephemeral state: - `actions`: Map of agent IDs to intended vector directions. - Observable information: Fully observable by all agents. ## Action space - `move(direction)`: `direction` ∈{NORTH,SOUTH,EAST,WEST,STAY}. - Preconditions: Only available to agents with `status == active` or `status == in_base`. Agents in `safe_return` execute an automated shortest-path return to base. ## Playability notes - Single-player-vs-AI: Multi-agent cooperative/competitive environment. A single human can control 1 agent with (K−1) cooperative teammate bots against K opponent bots. - Group games (n>4): Reduced naturally by setting K=2 or K=3 agents per team, preserving the complete decoy-tag-rescue strategic dynamic. - Physical-skill games: Physical sprinting and agility converted into spatial grid navigation and timestamp advantage calculation. - What is lost: Physical exhaustion, speed differentials, and diving tag reach. ## Implementation hints - Implement as a simultaneous-step multi-agent environment (Gym/PettingZoo compatible). - Key state update order per tick: (1) Apply desired moves → (2) Resolve boundary/base interactions → (3) Resolve hostile collision tags via timestamp comparison → (4) Resolve prison chain rescue checks → (5) Evaluate win conditions.

Download as JSON

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

Download JSON
Cite this page

Prisoner's Base (Greek Base Tag) (Greece). (n.d.). Games of Countries. https://gamesofcountries.com/games/gr-greece/ambariza

@misc{gamesofcountries-gr-greece-ambariza,
  title = {Prisoner's Base (Greek Base Tag) (Greece)},
  year = {n.d.},
  publisher = {Games of Countries},
  url = {https://gamesofcountries.com/games/gr-greece/ambariza}
}