Formal Specification
Central African Traditional Target Pitching Game — Formal Specification
Car · dexterity-throw · Confidence: low
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 2
# Ndölö — AI-optimized specification
> **Reconstructed source.** The entry carries a provenance warning: its rules were written from
> general knowledge, not read from the two works it cites. Samarin (1966) and Kalck (2004) are real
> books, but the entry does not evidence that either describes this game.
## General idea
A pitching game of the boules family, with the structural feature that distinguishes that family
from ordinary target games: **a thrown disc is both a scoring object and a tool for displacing
opponents' scoring objects**. Knocking a rival disc away from the stake is explicitly legal (R4),
so every throw carries two possible intentions and the second player to throw always holds an
option the first does not. Layered on top is a two-tier reward — nearest scores 1, touching the
stake scores 2 — which means a disc resting *on* the stake is worth attacking as well as beating.
Note the contrast with Kolo, the other Central African Republic pitching entry: that one scores by
removing tokens from a draining pool, this one by proximity to a fixed target, and they terminate
by entirely different arguments.
## Formal model
- **Abstraction taken:** Throwing accuracy over 5 m and disc-on-disc collision are collapsed into
stochastic outcomes — a landing-point draw and a displacement result. The entry specifies the
distance, disc budget, point values, and target score exactly, and supplies no physical model.
- **Player/agent count:** 2-6 (entry §Players).
- **Turn model:** Sequential throws within a round; scoring resolves once all discs are thrown.
## Formal rules
R1. Target (entry §1): a **stake** (*Kpou*) driven into packed earth **5 metres** from the throwing
line.
R2. Disc budget (entry §2): each player pitches **2 master discs** per round.
R3. Scoring (entry §3): the disc resting **closest to the stake** scores **1 point**; a disc
**directly touching the stake** scores **2 points**
[gap-fill: the entry does not say whether these are exclusive — whether a touching disc scores 2
instead of or in addition to being closest, nor whether multiple touching discs each score 2.
The two-tier reading is taken as exclusive: a touching disc scores 2, and if no disc touches,
the closest scores 1].
R4. Displacement (entry §4): **knocking an opponent's disc further from the stake is a legal
tactical move.** This is stated as permission rather than as a scoring event, so displacement
changes the geometry and affects R3's outcome indirectly rather than scoring on its own.
R5. Victory (entry §5): the first player to accumulate **12 points** wins.
R6. Throw order: unstated [gap-fill: the entry does not say whether the order is fixed, rotates
between rounds, or is determined by the previous round's result. Order matters greatly under
R4, since a later thrower can displace earlier discs and cannot be displaced in reply; a fixed
rotation is assumed].
## Edge cases
- Tie at the end: not applicable at the victory condition. R5 triggers the moment a total reaches
12 and points are awarded at the end of a round, so exactly one player crosses first — unless two
players reach 12 in the same round, which the entry does not address and which is possible only if
both scored that round. Within a round, two discs equidistant from the stake would tie R3's
proximity comparison; with continuous positions this is negligible, and no rule is invented.
- A player with no legal move: not applicable — every player has `pitch` available for each of
their 2 discs in every round, and discs are recovered between rounds.
- Repetition / cycles: rounds repeat until R5 is met. No cycle risk exists, because R3 awards at
least one point in every round to some player — which is exactly what bounds the game below.
- Simultaneous claims on one position: **R4 makes this the central mechanic rather than an edge
case.** Two discs cannot occupy the same point, and the resolution is physical displacement: a
disc arriving where another rests drives it away. The entry permits this explicitly and gives no
restitution, so the displaced player simply loses position.
- Elimination mid-round and turn-order repair: no player is ever eliminated, and discs are recovered
each round rather than lost. The throw rotation is fixed (R6's assumption) and needs no repair.
- Resource exhaustion: not applicable — the 2 discs per player are recovered every round and the
stake is permanent. Nothing depletes; the only bounded quantity is the score, which grows.
- Termination guarantee: **guaranteed deterministically.** Scores are monotone — nothing subtracts —
and R3 awards **at least one point in every round**, since some disc is always closest to the
stake. With `n` players, after at most `11n + 1` rounds some player must hold 12 points by the
pigeonhole principle. This is a hard bound rather than a probability-1 argument, and it holds
regardless of how the collisions in R4 resolve.
## State representation
- Persistent (match) state: `score[p]` (monotone non-decreasing, target 12), `current_round`,
`throw_order`.
- Per-round ephemeral state: `disc_positions[p][2]` in the plane, `discs_thrown[p]`,
`distance_to_stake` derived per disc, `touching[p][d]` (bool).
- Observable information: fully observable — an open clay court where the stake, every disc's
position, and all scores are visible. The one genuine informational asymmetry is temporal: an
early thrower commits without knowing what will follow, and a late thrower sees everything.
## Action space
- `pitch(aim, force, intent)`: the active player; precondition: they have discs remaining this
round (R2). The `intent` parameter is not cosmetic — under R4 a throw aimed at an opponent's disc
is a different action from one aimed at the stake, and an agent must choose between them.
- Note that unlike most dexterity entries in this corpus, this action space contains a **real
strategic decision** rather than a pure execution check, because R4 gives every throw two
purposes.
## Playability notes
- Single-player-vs-AI: genuinely good, and among the better dexterity entries here. R4 creates
interaction, R6's ordering creates positional advantage, and R3's two tiers create a risk gradient
— an agent must decide whether to defend a scoring disc, attack an opponent's, or go for the
2-point touch. That is a real decision problem rather than a skill measurement.
- Reducing n > 4 to 2 agents: survives well. With two players R4's displacement target is
unambiguous, and the alternating-throw structure preserves the game's core tension intact; what is
lost is the multi-way politics of choosing whose disc to attack.
- Physical-skill games: throwing accuracy and disc-on-disc collision are the entire physical
contest. Collision in particular deserves a real model rather than an outcome draw, since R4's
legality makes it a deliberate act whose geometry the player is trying to control.
- What is lost: the setting the entry records — packed clay courts, post-harvest gatherings among
Gbaya, Banda, and Sango-speaking communities — and the forest-encampment variant in which heavy
wild Canarium seeds serve as pitching tokens, which would change the collision dynamics
substantially.
## Implementation hints
Model disc positions in the plane and resolve R4's displacement as an actual collision, not as a
scripted "opponent's disc moves back" effect: the whole tactical layer depends on where a displaced
disc ends up, and a scripted displacement removes the risk that an attacking throw leaves your own
disc badly placed. Keep throw order explicit and visible, since R6's gap conceals a real advantage —
if the order is fixed across rounds, the last thrower is systematically favoured, and an
implementation should at least measure that before choosing. R3's exclusivity assumption should be
surfaced as configuration, because "closest scores 1, touching scores 2" admits a reading in which
both fire and that roughly doubles scoring rates. Do not borrow Kolo's token-removal scoring; the
two Central African Republic pitching entries end by different mechanisms and conflating them would
erase a real distinction. Suggested first milestone: a single round with planar disc positions, a
dispersion model, and elastic displacement, verifying that the pigeonhole bound holds — that every
round awards at least one point and matches always end.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Central African Traditional Target Pitching Game (Car). (n.d.). Games of Countries. https://gamesofcountries.com/games/cf-central-african-republic/ndolo-sango
@misc{gamesofcountries-cf-central-african-republic-ndolo-sango,
title = {Central African Traditional Target Pitching Game (Car)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/cf-central-african-republic/ndolo-sango}
}