Input: a CSV actual,predicted,group (1 = the station became/was predicted empty). Using scikit-learn (confusion_matrix(..., labels=[0, 1]), precision_score, recall_score, accuracy_score with zero_division=0), print:
confusion matrix (rows actual 0/1, columns predicted 0/1):
11 1
2 6
accuracy 0.85, precision 0.86, recall 0.75
central: n=10, positives=4, recall 1.00, accuracy 0.90
peripheral: n=10, positives=4, recall 0.50, accuracy 0.80
recall gap between groups: 0.50 <- the model misses far more problems in one group
(each matrix row: two spaces, then counts right-aligned in 3 characters separated by a space; groups alphabetical; the warning suffix only if the gap > 0.2).