CodeCup 2027 Update

Continuing with the CodeCup 2027 posts, this update reports the results of the 5th training round and an update to the training evaluations.

5th Training Round Results

Points	Player
35478	pnn5
33280	pnn4
32735	pnn3
27764	pnn2
26655	pnn1
23366	player3
21460	player2
18222	player1

pnn5 finished ahead of pnn4 by 2,198 points — a wider gap than pnn4's 545-point lead over pnn3.

Training Evaluations Update

While writing the previous post, I had AI review the post for any errors or issues. In the post, I mentioned that the pnn4 player showed a minor improvement over the pnn3 player. The AI noted that when the pnn3 and pnn4 players played each other, they tied on the win-loss record.

While the 5th training round was underway, I informed the AI that the scores were added up for each player in the competition and the ranking in the competition was determined by the total score, which is slightly different from a win-loss record used in the training rounds.

Fable 5 suggested updating the training to focus on the total score rather than the win-loss record. This update was completed, and there wasn't enough time to complete the 6th training round before the test competition on Saturday. It was taking about 3 hours to train each generation, and training 10 generations would take too long.

The update to the training will be used for all training rounds going forward.

Scoring

The scoring of a game is as follows:

1. Players score points based on the territory they control.
2. The winner gets a bonus of 200 points.
3. The loser gets a bonus of 100 points.
4. A program that makes an invalid move, crashes, exits early, or otherwise behaves incorrectly gets a score of 0 points, regardless of territory. In the competition, if a program fails in one of those ways, it gets replaced by another program so that the other player can still compete and earn points.

Fable 5 points out that the win-loss record is embedded in the scoring system in the 100-point difference between the winner and loser bonuses.