Formal Specification
Beti-Fang Carved Nutshell Gambling Game — Formal Specification
Cameroon · dice-lots · Confidence: low
← Back to game pageImplementation Suitability
Suitability: engine-readyFidelity: directGap-fills: 4
# Abbia — 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. Tessmann (1913) and Laburthe-Tolra
> (1981) are real monographs, but the entry does not evidence that either describes these rules.
> **Wagering game.** The mechanic is staking and payout. An implementation aimed at a general
> audience should represent the stakes as abstract tokens, which is how the entry itself describes
> them.
## General idea
A four-shell lots game whose randomness is **genuine rather than proxied** — unlike almost
everything else in this corpus, the uncertainty here is the actual mechanic and not an abstraction
of physical skill. Four carved shells are shaken in a raffia basket and flipped onto a mat; the
count of carved faces showing determines the payout. There are no decisions at all once the stake
is placed: the caster shakes, the faces land, the pot moves. What makes the file worth close
reading is that the payout structure, taken at face value with fair shells, is **sharply negative
for the caster** — see the note under R4, which is a property of the entry as written rather than a
claim about how the game was actually played.
## Formal model
- **Abstraction taken:** None of consequence. Shaking and flipping produce a discrete random
outcome over 4 binary faces, which is exactly what the model represents. The one abstraction is
the assumption of **fair, independent faces**, which the entry neither states nor contradicts.
- **Player/agent count:** 2-8 adults (entry §Players).
- **Turn model:** Strictly sequential; the basket passes clockwise after each cast.
## Formal rules
R1. Staking (entry §1): each player places an **agreed stake** of tally tokens into the central pot
[gap-fill: the entry does not fix the stake size, whether it is uniform across players, or
whether players may decline to stake in a given round].
R2. Loading (entry §2): the active caster places **4 carved Abbia shells** and **4 matching
counting discs** into the woven basket (*Mbi-nza*)
[gap-fill: the counting discs are named and never used again — the entry assigns them no role
in resolution. They are excluded from the outcome model, and their purpose is unrecorded].
R3. Cast (entry §3): the basket is shaken vigorously and flipped onto a woven mat. Each shell
lands **carved (convex) face up** or **plain (concave) face up**.
R4. Payout (entry §4), by the count of carved faces showing:
- **4 carved — *Mfem*:** highest winning roll; the **caster sweeps the pot**.
- **4 plain — *Nkik*:** second winning roll; **double return** to the caster.
- **2 carved, 2 plain — *Koba*:** neutral; bets stay in the pot.
- **3-1 split — *Samba*:** losing roll; the **caster pays all active bettors**.
Under fair independent faces these have probabilities 1/16, 1/16, 6/16 and **8/16**
respectively — so the caster loses outright on half of all casts and wins on one eighth
[gap-fill: the entry gives no payout multipliers beyond "sweeps" and "double return", and no
stake size, so the expected value cannot be computed and the asymmetry cannot be checked. Real
Abbia shells are carved on one face and may well not be fair; the entry says nothing, so
fairness is assumed and flagged rather than asserted].
R5. Rotation (entry §5): the basket passes **clockwise** to the next player, who becomes the
caster.
R6. Match structure: **none.** The entry defines no number of rounds, no bankruptcy rule, and no
end condition [gap-fill: whether a player who loses all tokens leaves the game is unstated, so
nothing terminates the rotation].
## Edge cases
- Tie at the end: not applicable — there is no end (R6) and no score, only token holdings that
change hands. Nothing is ever compared, so nothing can tie.
- A player with no legal move: **reachable and unaddressed.** A player whose tokens are exhausted
cannot meet R1's stake, and the entry provides no rule for it — no exit, no credit, no minimum.
This is the entry's sharpest omission, because it is the natural end of a wagering game and the
entry does not use it.
- Repetition / cycles: the rotation under R5 continues indefinitely by construction. Every cast is
independent, so the game is a stationary random walk in token holdings with no absorbing state
defined — which is precisely why termination fails below.
- Simultaneous claims on one position: not applicable — one caster at a time (R5), one pot, and a
single outcome per cast. The four shells resolve as one joint event, not as four competing ones.
- Elimination mid-round and turn-order repair: **no elimination is defined**, though R1 makes it
the obvious consequence of running out of tokens. If a bankrupt player were removed, the
clockwise rotation of R5 would simply skip them and need no repair — but the entry does not say
they are removed.
- Resource exhaustion: **the central unmodelled dynamic.** Tokens are conserved overall — they move
between players and the pot rather than being consumed — but an individual player's holding is a
random walk that can reach zero. The entry tracks the pot and never the players' reserves.
- Termination guarantee: **not guaranteed.** Nothing bounds the game: no round count (R6), no
target, and no bankruptcy exit. Token totals are conserved rather than monotone, so no resource
drains to zero at the table level. Under a fair-shell model a given player is eventually ruined
with probability 1 in an unbounded walk, but the entry defines no consequence for ruin, so even
that does not end play. `termination_guaranteed: false` records genuine silence, and an
implementation must supply both a bankruptcy rule and a stopping condition, labelled as its own.
## State representation
- Persistent (session) state: `tokens[p]` (a conserved but individually unbounded quantity),
`pot`, `caster_index` (advancing clockwise per R5).
- Per-cast ephemeral state: `faces[4]` ∈ {carved, plain}, and the derived `carved_count` ∈ {0..4}
which is the only thing R4 actually reads.
- Observable information: fully observable. The cast is public, the pot is public, and there is no
hidden state — no concealed hand, no private information, and no bluffing surface. This is worth
stating because it is unusual for a wagering game and means there is nothing to model about
opponents.
## Action space
- `stake(amount)`: each player before a cast (R1). This is the **only** decision in the game, and
the entry does not even establish that the amount is free rather than fixed.
- `cast()`: the active caster; resolves R3 and R4 automatically with no parameters.
- Note the consequence: with `stake` fixed by agreement, the action space is empty of real choices
and the game is a pure random process observed by its participants.
## Playability notes
- Single-player-vs-AI: pointless as an adversarial problem, since no decision separates a good
player from a bad one. There is nothing to learn, nothing to exploit, and no opponent model worth
building — an AI participant is a random number generator with a name.
- Reducing n > 4 to 2 agents: mechanically lossless, because players do not interact beyond the
shared pot. What shrinks is the pot size and the social scale, which is most of the point.
- Physical-skill games: not applicable in the usual sense, but note the inversion — where most
entries in this corpus abstract skill into randomness, this one **is** randomness, so the model
loses nothing. The shaking technique might in reality bias the outcome, and the entry does not
suggest it does.
- What is lost: everything the entry says makes Abbia significant. The shells carry miniature
carved iconography — pangolins, leopards, hornbills, scenes of Beti village life, ancestral
proverbs — preserved in ethnographic museums worldwide. The game's cultural weight is in the
**carving**, and the formal model reduces each shell to a single bit.
## Implementation hints
Model the outcome as a draw of `carved_count` from a binomial distribution over 4 shells, with the
per-shell probability exposed as a parameter rather than hard-coded at 0.5 — this is the one knob
that decides whether R4's payout table is playable, and the entry does not fix it. Surface the
caster's expected value as a diagnostic during development: under fair shells the caster loses on
8 of 16 outcomes and wins on 2, and if a simulation shows the caster's holdings falling steadily,
that is the entry's payout table behaving as written rather than a bug. R2's unused counting discs
should be represented as what they are — an unexplained element — not quietly repurposed as a
scoring mechanism. Both missing bounds, the bankruptcy rule and the stopping condition, must be
added explicitly and labelled. Suggested first milestone: the cast-and-payout loop with
configurable face probability and token accounting, verifying conservation (all tokens plus the pot
is constant) and reporting how quickly a player is ruined.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Beti-Fang Carved Nutshell Gambling Game (Cameroon). (n.d.). Games of Countries. https://gamesofcountries.com/games/cm-cameroon/abbia-cameroon
@misc{gamesofcountries-cm-cameroon-abbia-cameroon,
title = {Beti-Fang Carved Nutshell Gambling Game (Cameroon)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/cm-cameroon/abbia-cameroon}
}