Formal Specification
Snatch the Coin / Coin Grabbing from Horseback — Formal Specification
Kazakhstan · equestrian-team · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 5
# Tenge ilu — AI-optimized specification
## General idea
Tenge ilu is a high-speed equestrian dexterity and gymnastics race in which a mounted rider sprints down a 150-metre track at full gallop, leaning steeply from the saddle to snatch cloth-wrapped coin pouches from ground pits within a strict time ceiling. Agents must balance aggressive reach depth against saddle recovery and horse speed, since breaking into a trot or exceeding the time limit forfeits all points.
## Formal model
- **Abstraction taken:** The continuous 3D biomechanics of horseback leaning (stirrup torque, core extension, ground clearance, and reins control) are modeled as a sequence of discrete coin-station decision points along a timed galloping run. [gap-fill] At each station , the agent selects an effort intensity , trading off probability of pouch capture against recovery stability and gait-break penalty risk.
- **Player/agent count:** 1 to agents competing in sequential timed heats (standard 2-agent tournament match over 3 runs).
- **Turn model:** Sequential single-agent timed runs; each run resolves a multi-stage discrete traversal.
## Formal rules
R1. **Track Layout and Course Parameters:**
- Total track length: 150 metres.
- Time ceiling: seconds.
- [gap-fill] Exactly 5 coin stations () located at distances 40m, 60m, 80m, 100m, and 120m from the starting line.
- Each station contains 1 coin pouch worth 1 point.
R2. **The Timed Run Traversal:**
- For each station , the agent chooses effort parameter .
- **Capture Probability:** .
- **Stability and Gait Check:**
- If , there is a risk of losing balance or disrupting horse stride: .
- If a gait break occurs (horse drops to trot or stops): the run is immediately invalidated, resulting in 0 points for that heat.
- **Elapsed Time Computation:**
- [gap-fill] Base run duration , where .
- If total elapsed time (16.0s), the run is disqualified (0 points).
R3. **Heat Scoring:**
- If the run is valid (no gait break and ):
- .
- .
- If disqualified: , .
R4. **Match Progression and Victory:**
- A match consists of 3 regulation runs per agent.
- Total match score = .
- Total match time = .
- The agent with the highest total match score wins.
- If total scores are tied, the agent with the strictly lower total elapsed time wins.
- [gap-fill] If both score and time are tied, a sudden-death 1-run shootout is conducted.
## Edge cases
- **Tie at the end condition:** Explicitly resolved by R4 (fastest cumulative time breaks score ties; exact ties enter sudden-death shootout [gap-fill]).
- **A player with no legal move:** Not applicable; an agent always selects effort vectors from for each scheduled run.
- **Repetition / cycles (and the loop-breaking rule):** Runs are strictly finite in distance (150m) and time (), with exactly 3 runs per match.
- **Simultaneous claims on the same resource or position:** Not applicable; agents run the course in strict isolation one at a time.
- **Player elimination mid-round; turn order repair:** If an agent is disqualified in a heat, they receive 0 points for that heat and proceed to subsequent heats; if an agent withdraws, remaining agents finish their heats.
- **Resource exhaustion (pieces, tokens, space, time):** Coin stations are freshly restocked before every individual run.
- **Termination guarantee: why the game provably ends:** A match is defined over exactly 3 discrete, bounded runs per agent. Bounded finite execution guarantees immediate termination.
## State representation
- **Persistent state:**
- `agent_scores`: Array of integers (cumulative pouches).
- `agent_times`: Array of floats (cumulative elapsed seconds).
- `current_heat`: Integer .
- `active_agent_idx`: Integer .
- **Per-run ephemeral state:**
- `station_results`: Array of 5 booleans `[snatch_1..snatch_5]`.
- `gait_broken`: Boolean flag.
- `elapsed_time`: Float.
- **Observable information:** Fully observable. Target stations and historical run scores/times are public knowledge.
## Action space
- `execute_run(effort_vector)`:
- Parameters: `effort_vector` with each .
- Preconditions: Active agent's scheduled turn in current heat.
## Playability notes
- **Single-player-vs-AI:** Natural. 1 human vs 1 AI agent or leaderboard of ghost times.
- **Group games (n>4):** Scales trivially to any tournament size since runs are independent sequential time trials.
- **Physical-skill games:** Skill modeled via two parameters: (snatch success) and (stride stability / gait maintenance).
- **What is lost in the adaptation:** Real-time physical sensations of saddle tilt, wind shear, horse stride timing, and fingertip grazing on the turf.
## Implementation hints
- Represent each run as a vectorized Monte Carlo sample given the agent's 5-element decision vector.
- The optimization problem has an intuitive trade-off: higher effort increases pickup chance but increases run time towards the 16.0s disqualification cliff.
- First milestone: Implement single-run stochastic simulation with deterministic time formulas, verify score/time ranking, and validate tie-break logic.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Snatch the Coin / Coin Grabbing from Horseback (Kazakhstan). (n.d.). Games of Countries. https://gamesofcountries.com/games/kz-kazakhstan/tenge-ilu
@misc{gamesofcountries-kz-kazakhstan-tenge-ilu,
title = {Snatch the Coin / Coin Grabbing from Horseback (Kazakhstan)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/kz-kazakhstan/tenge-ilu}
}