Research

We extracted 500 boards from the original Apple Game using screenshots and OCR, then statistically analyzed the board generation algorithm. After examining 85,000 digits, we found that each cell is drawn independently and uniformly from 1 to 9, with no evidence of positional bias, adjacent-cell correlation, or digit clustering. We confirmed that the original game uses Rejection Sampling, and boards generated with our implementation showed no significant difference from the original data (KS test, p=0.65).

We analyzed 500 boards from the original Apple Game to find a simple metric that can estimate board difficulty. The total sum of board numbers best predicts difficulty (r = -0.751), explaining 56% of score variance. Based on this, we assign a 6-level difficulty rating. Among individual digits, count of 9 has the strongest negative correlation since 9 can only pair with 1.