Results ARC-AGI-2 Public training subset (14/120 tasks): 11.67% Public evaluation: 0% Private evaluation on Kaggle: 0%