Formal Specification
Aunt Sally (Pub Throwing Game) — Formal Specification
United Kingdom · dexterity-throw · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: probabilistic-proxyGap-fills: 2
# Aunt Sally — AI-optimized specification
## General idea
A turn-based throwing game where agents cast 6 wooden sticks per leg at a small doll perched on a metal swivel 30 feet away. The defining mechanic is the strict collision condition: striking the supporting iron armature or swivel before or during contact with the doll invalidates the score, creating a narrow target corridor that penalizes low or off-center throws.
## Formal model
- **Abstraction taken:** Physical stick trajectory (rotation, velocity, angle) is projected onto a 2D impact plane at distance 30 ft. The target area comprises three collision zones:
1. `Clean Doll Zone` (upper region, diameter 3.25 in, height 6 in) -> 1 Point.
2. `Iron / Swivel Foul Zone` (supporting stem below and behind doll) -> 0 Points (Foul / "Iron").
3. `Miss Zone` (outside target envelope) -> 0 Points.
- **Player/agent count:** 2 agents (in 1v1 match) or 2 teams of 8 agents.
- **Turn model:** Sequential turns: in each leg, Agent A throws 6 sticks consecutively, followed by Agent B throwing 6 sticks.
## Formal rules
R1. **Match Structure:** A match consists of the best of 3 legs (first to win 2 legs).
R2. **Leg Execution:** On their turn, an agent executes 6 consecutive throw actions:
- `throw_stick(aim_y: float, aim_z: float, spin_rate: float)`
R3. **Collision Sampling:** The thrown stick produces an impact point `(y_impact, z_impact)` sampled from a 2D Gaussian centered on `(aim_y, aim_z)` with variance `sigma` inversely proportional to agent accuracy skill:
- If impact falls within `Doll_Radius` AND above the `Iron_Collar_Threshold`: Score increments by 1 ("Doll").
- If impact contacts the `Iron_Collar` or `Swivel_Arm`: Hit is ruled "Iron", score = 0.
- If impact misses all structures: score = 0.
R4. **Leg Resolution:** Total dolls scored across the 6 throws (or 48 throws in team mode) are compared. The agent/team with the strictly higher count wins the leg.
R5. [gap-fill] Tie-Break (Tie in Leg):** If both agents finish a leg with equal score, a 3-stick sudden-death throw-off occurs. If still tied, single-stick sudden-death rounds repeat until broken.
R6. **Match Victory:** The first agent/team to secure 2 legs (in best-of-3) wins the match.
## Edge cases
- **Tie at the end condition:** Handled by R5 via multi-tier sudden-death throw-offs (3 sticks, then 1 stick) with non-zero probability of differential hit.
- **A player with no legal move:** Not applicable; a throw is always legal.
- **Repetition / cycles:** Bounded by leg limits and strictly decaying probability of infinite tied sudden-death rounds.
- **Simultaneous claims on the same resource:** Not applicable; throws occur strictly sequentially.
- **Player elimination mid-round:** In team format, if an agent is disqualified or disconnects, [gap-fill] remaining throws for that slot count as 0 dolls.
- **Resource exhaustion:** Exactly 6 sticks per player per leg; reset after each leg.
- **Termination guarantee:** Match termination is guaranteed by the finite number of legs and independent non-identical throw probabilities during tie-break resolution.
## State representation
- **Persistent state:** `legs_won: [int, int]`, `current_leg: int`, `team_scores_current_leg: [int, int]`, `active_agent_index: int`.
- **Per-throw ephemeral state:** Throws remaining in current turn `throws_left in [0..6]`, last impact coordinates `(y, z)`, outcome classification (`DOLL`, `IRON`, `MISS`).
- **Observable information:** Fully transparent public scoreboard and throw history.
## Action space
- `throw_stick(aim_y, aim_z)`: 2D continuous aim coordinates targeting the doll relative to the swivel center.
## Playability notes
- **Single-player-vs-AI:** Highly natural 1v1 game; players calibrate aim offsets to avoid clipping the lower iron collar while maximizing probability of hitting the doll center.
- **Group games reduction:** Seamlessly scales between 1v1 singles and 8v8 pub teams without altering core throwing mechanics.
- **Physical-skill games:** Modeled via the 2D Gaussian error distribution around the chosen aim coordinate.
- **What is lost in the adaptation:** Aerodynamic gyroscopic stick rotation and the acoustic "clack" feedback distinguishing wood-on-wood from wood-on-iron.
## Implementation hints
- Very low state footprint (few integer counters).
- Branching factor is 2D continuous aim (can be discretized to a 10x10 target grid).
- First milestone: Construct the 2D collision discriminator with doll, iron, and miss zones, and simulate 1,000 matches between agents of varying skill parameters.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Aunt Sally (Pub Throwing Game) (United Kingdom). (n.d.). Games of Countries. https://gamesofcountries.com/games/gb-united-kingdom/aunt-sally
@misc{gamesofcountries-gb-united-kingdom-aunt-sally,
title = {Aunt Sally (Pub Throwing Game) (United Kingdom)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/gb-united-kingdom/aunt-sally}
}