Formal Specification
Stick-Catching Circle Game — Formal Specification
New Zealand · dexterity-throw · Confidence: medium
← Back to game pageImplementation Suitability
Suitability: adaptation-requiredFidelity: probabilistic-proxyGap-fills: 9
# Poi Rākau — AI-optimized specification
## General idea
One agent stands at the centre of a ring and must intercept staves thrown at it by
the ring. Structurally it is a single-target interception problem under a throw
schedule the catcher does not control — closer to a reaction benchmark than to a
strategy game, and it is honest to say that most of what makes it a game is missing
from the available sources.
## Formal model
- **Abstraction taken:** a throw is replaced by an incoming-object event with a
time-to-arrival and a bearing; the catch is a probability check against the
catcher's reaction budget. No flight is simulated.
- **Player/agent count:** 1 catcher, n throwers with n ≥ 3.
- **Turn model:** ticked. Throws are scheduled onto ticks; the catcher reacts within
the tick.
## Formal rules
R1. The catcher occupies the centre; throwers occupy fixed bearings on a circle.
R2. On each tick a subset of throwers releases a rākau toward the catcher.
R3. [gap-fill] Exactly one rākau is in flight at a time, released in sequence
around the circle. The entry marks throw order and simultaneous count
undocumented; sequencing is the minimum assumption that makes the game
computable.
R4. A catch succeeds with probability `p_catch(reaction, bearing_change)`, decreasing
in how far the new bearing is from the previous one. [gap-fill] No source gives
a catching model.
R5. [gap-fill] A drop costs the catcher one life; the catcher starts with three.
R6. [gap-fill] When the catcher's lives are exhausted, the role passes to the
thrower whose throw was dropped last. The entry marks role transfer undocumented.
R7. [gap-fill] No score is kept. The entry states scoring is undocumented, so the
spec tracks catches purely as a statistic, not as a win condition.
R8. [gap-fill] The session ends after every player has held the centre once.
## Edge cases
- **Tie at the end condition:** with no score there is no tie to resolve. If catches
are used as an informal score, equal counts are left tied — the sources support no
tie-break.
- **Player with no legal move:** the catcher always has the same single action
available, and throwers always have a throw available, so no agent is ever stuck.
- **Repetition / cycles:** the throw sequence cycles the ring indefinitely. R5 and R8
are what bound it; without those gap-fills the activity has no end state at all.
- **Simultaneous claims:** R3 removes contention by allowing only one rākau in flight.
Were simultaneous throws permitted, the catcher would need an explicit priority
rule, which no source provides.
- **Mid-round elimination and turn-order repair:** nobody is eliminated; the centre
role transfers under R6 and the ring closes by index order.
- **Resource exhaustion:** catcher lives (R5) are the only consumable, and their
exhaustion triggers the role transfer.
- **Termination guarantee:** each centre tenure ends after at most three drops, and
R8 bounds the number of tenures to the player count, so the session provably ends.
Note that both bounds are [gap-fill] — the attested game has no termination rule.
## State representation
Persistent: ring bearings, player count.
Per-player: role (`catcher|thrower`), bearing, whether they have held the centre.
Per-turn ephemeral: active rākau bearing, time-to-arrival, catcher lives remaining,
previous bearing (needed for R4).
Information set: fully observable. The catcher sees the ring and the release; the
difficulty is reaction time, not concealment.
## Action space
- `throw(from_bearing)` — precondition: actor is a thrower, it is the actor's turn in
the ring sequence, no rākau is in flight.
- `catch()` — precondition: actor is the catcher and a rākau is in flight.
## Playability notes
- **Single-player-vs-AI:** natural in the sense that the catcher role is a
single-agent reaction task. But there is little to decide — the catcher's only
action is to attempt the catch — so it is a benchmark rather than a game.
- **Group games (n>4):** reduces to one catcher and one thrower without losing the
interception problem, though it loses the bearing-change difficulty that R4 models,
which is the main source of variance.
- **Physical-skill model:** one parameter, catch probability as a function of how far
the catcher must swing between successive bearings.
- **What is lost:** almost everything. Six of eight rules are [gap-fill]. Strip
them and what the sources actually attest is: people stand in a circle and throw
staves at a person in the middle, who catches them. An implementer should treat
this spec as scaffolding awaiting real documentation, not as a faithful model.
## Implementation hints
Do not invest in flight simulation; the entire model is a bearing sequence and one
probability. The representational trap is treating the gap-filled scoring and lives
rules as attested — they are not, and an implementer who wants fidelity should strip
every [gap-fill] first and see how little remains. First milestone: a ring, a
throw sequence, and a catch check.
Download as JSON
Download the full game corpus entry for this game as structured JSON.
Download JSONCite this page
Stick-Catching Circle Game (New Zealand). (n.d.). Games of Countries. https://gamesofcountries.com/games/nz-new-zealand/poi-rakau
@misc{gamesofcountries-nz-new-zealand-poi-rakau,
title = {Stick-Catching Circle Game (New Zealand)},
year = {n.d.},
publisher = {Games of Countries},
url = {https://gamesofcountries.com/games/nz-new-zealand/poi-rakau}
}