- Uses hardest difficulty
- Aims for maximum score
- Genre: Board
ROM Choice: Two ROM variants exist. The one used in this TAS is the later variant and is easily distinguished by the four dots at the corners of the center 4x4 grid. I decided on this ROM as it is by far the most common one and likely the one people will be more familiar with.
Difficulty Choice: Game 3 (Expert) is the hardest difficulty and was chosen here. However even on expert difficulty, this game really doesn't pose much of a challenge to an experienced player, which I am NOT.
Goal and Methodology
An Othello submission was rejected in the past for having too trivial a goal so I wanted to go for something bigger here: find a variation where the player wins 64-0. Some examples of people getting such a score in real time can be found on Twin Galaxies here and here. In testing routes for this TAS, I replicated the sequence of moves made by the four 64-0 games on Twin Galaxies but taking the absolute minimum time to select the next move. Since I am a terrible Othello player and could never win a game 64-0 myself, I also used the Othello computer program WZebra to play a whole bunch of games against the Atari 2600 and see if it could win 64-0.
The WZebra settings I initially used for testing included a search depth of 24 moves + last 26 perfect, play human and computer-style openings, no opening book or midgame randomness, and use a 256MB position table. I ran games with these settings until I started to get duplicate games. By the time that happened, I got no games that resulted in a 64-0 win. None. At that point, I progressively increased the opening book and midgame randomness until I got a couple games that ended 64-0. I then trimmed these games down to spend the minimum time taking a turn.
For all the valid Twin Galaxies and WZebra routes found, I also looked at the last couple of moves manually to see if there were any obvious move order changes that could reduce the number of checkers flipped, the number of cursor movements, or just end input earlier to make sure they were all as fast as they could be before comparing with each other.
There is some computer randomness—not a lot—during the opening and midgame phases. By delaying when you place your checker, the computer may place theirs in a different place on the next turn, provided it thinks more than one move are about equal. This was used in some places to replicate the moves played in the Twin Galaxies and WZebra games for TASing purposes.
Results and Postprocessing
The results were quite interesting. The fastest route ended up being one of the Twin Galaxies runs by ZilchSR and not one of the computer games. So congrats to him for putting most of that route together! I changed the last nine moves so the long row of six checkers on the edge would be the last checkers flipped—great for ending input early. I then realized that ZilchSR's game starts with a checker being placed on c4 which is four cursor units away from the starting point. If we assume a 180-degree rotation of the board, that checker would be placed at f5, three cursor units from the starting place. So I redid the entire game rotated 180 degrees to get out one less cursor move; but I got worse luck so it only ended up saving 3 frames. I then realized I could flip the already rotated board on the upperleft-lowerright diagonal axis so the first checker could be placed at e6 which is still three cursor units from the start. After redoing the entire game again with this board transformation, I got much better luck and saved an additional 9 frames.
feos: Claiming for judging.