The Awari Oracle was a web frontend of a remote database containing the eventual outcome of all the 889,063,398,406 positions that can occur during a game of Awari, a variant of Oware.
The Awari rules were invented by the computer scientists Victor Allis, Maarten van der Meulen, and H. Jaap van den Herik in 1991.
They differ from Oware as follows:
- A position that is repeated for the third time (the Oracle stops after the second repetition) results in a even division of the remaining stones, including a possible odd stone.
- It is not allowed to do a move that leaves the opponent without countermove, unless all moves eradicate the opponent and at least one seed is left on the player's own side. If the opponent is eradicated by a capture the remaining stones are awarded to the player on whose side they are.
The first rule is only used by computer scientists, but not by human players. It can lead to strange results such as 23.5 vs 24.5 seeds. The second rule is used in some Oware variants, but not in tournaments. The international rules (abapa) permit a player to do such a move, but then he captures nothing.
The Oracle is a Java applet, which was developed by John W. Romein and Henri E. Bal at the Division of Mathematics and Computer Science, Faculty of Sciences, Vrije Universiteit Amsterdam, in 2002, as part of their research to solve Awari. It can be used in two modes, both needing a working connection to the database, game mode and edit mode.
Users can set the mode to "game" and then play against the applet. They can choose the playing level between fair, good, strong, expert and perfect. There is no information about how the Oracle chooses its moves when in the first four levels, but in the fifth level (perfect) it plays according to the results stored in the database, i.e. perfectly.
In edit mode, the users choose a position and then get the eventual outcome of the game for every possible move, if played perfectly from then on.
The database was created using parallel retrograde analysis to get the eventual score of all possible positions, if both players would play perfectly from that position on. A computer cluster was needed to compute it, requiring 51 hours. The grid (connected through Myrinet) was made of 72 dual 1GHZ PIII computers, each with 1GB main memory.
The database requires 778GB for being stored.
Romein and Bal proved in 2002 that the game-theoretic value of Awari is a draw. The only perfect move to start a game is pit 6 (the last pit). The following move can be pit 2, 5 or 6. All the other moves are losing.
Testing Human Players
The Oracle was used to test the strength of Oware playing programs such as the Oware Wizard. In 2002, the Catalan Oware expert Víktor Bautista i Roca demonstrated that more than 95% of all moves in a game between two master players, Trevor Simon and Ibrahim Abubakar, were perfect according to the Oracle. However, a final conclusion cannot be drawn as the data are based on rules slightly different from Oware.
Due to technical problems (the server crashed and was difficult to revive) the Oracle was offline from September 2007 to August 2008. The database was regenerated, using a wide-area network, and installed on a different server by Kees Verstoep.
In 2014, the Awari Oracle became defunct again.
Víktor Bautista i Roca claimed on his now defunct homepage manqala.org that several Awari endgames were incorrectly analyzed by the Awari Oracle. Therefore it seems that Romein and Bal's research had major flaws and that their results are not valid. However, both sites (manqala.org and the Oracle) have been taken off the internet and no further research appears to be possible. This reveals a major problem: Most research done in solving games is not really peer-reviewed. Minor mistakes in the programming, which nevertheless can give quite different results, will usually go unnoticed.
- Allis, L. V., Meulen M. van der & Herik, H. J. van den
- Databases in Awari. In: Levy, D. N. L. & Beal, D. F. (Eds.). Heuristic Programming in Artificial Intelligence 2. The Second Computer Olympiad. Universiteit Maastricht, Maastricht (Netherlands) 1991, 73-86.
- Donkers, J.
- Comments on the Awari Solution. In: ICGA Journal 2002; 25 (3).
- Romein, J. W. & Bal, H. E.
- Awari is Solved. In: ICGA Journal 2002; 25 (3).
- Romein, J. W. & Bal, H. E.
- Solving the Game of Awari using Parallel Retrograde Analysis. In: IEEE Computer 2003; 36 (10): 26-33.
- Verstoep, K., Maassen J., Bal H. E. & Romein, J. W.
- Experiences with Fine-grained Distributed Supercomputing on a 10G Testbed.. In: 8th IEEE International Symposium on Cluster Computing and the Grid. CCGRID '08, 19-22 May 2008, 376-383.