NAME
Game::Oware::Variant - the rule table, and what varies between rule sets
VERSION
Version 0.01
SYNOPSIS
use Game::Oware::Variant qw(grand_slam target);
grand_slam('abapa'); # no_capture
grand_slam('awari'); # illegal_unless_only
target('abapa'); # 25DESCRIPTION
A table of named rule flags with one accessor per field, in the shape of Game::Schnapsen::Variant. Not subclasses: two rule sets that differ in five scalars do not need two class hierarchies, and a spec table is the only form in which the difference can be read at a glance.
abapa is the default, and the only one a player will meet
The rules in the Wikipedia article's own body, "considered to be the most appropriate for serious, adult play". Everything else here exists for the test suite.
grand_slam names the policy, and the two shipped are not both Wikipedia's
A grand slam is a move whose capture chain would take every seed the opponent has. Wikipedia lists four readings of it. This distribution implements two policies, and only one of them is on that list:
no_capture-
The move is legal and captures nothing. Shipped as
abapa, because it is what the article's own rules body already describes - "the capture is forfeited ... and the seeds are instead left on the board" - and because the Variations section says international competitions follow it. This is Wikipedia's variation 2. illegal_unless_only-
The move is not legal at all, unless it is the only move available, in which case it is played and captures normally. Shipped as
awari, and this is the rule set Romein and Bal solved.
Where the awari rule set comes from, and why it is not what Wikipedia says
awari exists for one reason: the Awari Game Score Database published by the Vrije Universiteit (DOI 10.48338/vu01-11wjke, CC-BY 4.0) is an exact per-position oracle, and it is an oracle for the rule set they solved rather than for the one the site plays.
Wikipedia is not a source for that rule set. It describes it only as "variations allowing Grand slams to end the game", a phrase matching none of its own four numbered variations, and taking it at face value would have shipped "the capture is made and the rest of the board goes to the opponent", which is its variation 3 and is wrong.
The rule is pinned instead from the dataset's own Awari-Python/README.md, fetched 18 September 2026, which says of the scores it returns:
The score printed is the score of the current board interpreted
according to the special rules regarding which moves are acceptable.
Specifically, it is not allowed to remove all stones of the opponent
(leaving it no move), unless it is the only move available.Note what that single sentence does: it is Wikipedia's variation 1 with an exception clause, and it subsumes the feeding rule. "Leaving the opponent no move" covers both capturing all their seeds and failing to give seeds to an opponent who has none, so awari needs one rule where abapa needs two. The exception clause is what keeps a position with no other option playable.
The dataset pins nothing else. Its README.md, its description and its Board.py cover sowing, capture and Goedel indexing only; Board.py has no move generation at all. So the cycle rule and the treatment of a position where neither side can move are still unpinned, and the other four fields here are copied from abapa. An oracle test has to keep away from positions where that matters, and xt/awari-oracle.t says so.
plies_without_capture is measured, and it is a backstop rather than the rule
Five hundred. It is a house rule with no published number behind it, so the value is measured rather than chosen, and xt/cycle-measure.t is the measurement.
Over three hundred bot games at mixed levels, with the cap lifted so the sample is not truncated by the thing it is meant to justify, the longest capture-free stretch was 396 plies, the 95th percentile was 124 and the median was 19. Five hundred is above the observed maximum with margin, and well inside the scale of a game, so it ends a game that has genuinely stopped progressing and never one that was still going somewhere.
The development placeholder was fifty, which the same measurement shows would have fired in well over a third of ordinary games.
At five hundred this trigger is the outer bound and repetition is what actually ends cyclic games: in that sample, threefold repetition ended 24 games of 300 and the ply cap ended none. That is the division of labour D2 described, and an earlier reading of the evidence had it backwards because it was drawn from games played by a first-legal-move policy rather than by a bot.
FUNCTIONS
Nothing is exported by default.
variants
Every variant name, sorted.
fields
Every field name, sorted. The two specs carry the same keys, and this is what proves it.
is_variant
True for a known variant name.
check_variant
Returns its argument, or dies. An unknown variant is programmer error.
spec_for
A copy of one variant's whole table, so a caller cannot edit the original.
grand_slam
no_capture or illegal_unless_only.
target
The seeds that win a game. Twenty-five.
draw_at
The seeds each that draw one. Twenty-four.
repetition_limit
How many times a position may occur before the cycle rule ends the game.
plies_without_capture
How many plies may pass with no seed entering a store before the cycle rule ends the game. Five hundred, measured rather than chosen; see above.
SEE ALSO
Game::Oware, Game::Oware::Rules
AUTHOR
LNATION <email@lnation.org>
LICENSE AND COPYRIGHT
This software is Copyright (c) 2026 by LNATION <email@lnation.org>.
This is free software, licensed under the Artistic License 2.0.