Moderation threshold tuning

You are tuning the decision threshold for a content-moderation model that outputs integer scores 0-100. At threshold t, an item is predicted 1 when its score >= t, else 0. Given scores, binary labels, and a list of candidate thresholds, return the candidate that maximizes the number of correct predictions. If several candidates tie on correct count, return the smallest such threshold value. Candidates may be unsorted. If candidates is empty, return -1. Example: scores = [10, 80, 55, 95, 30, 70], labels = [0, 1, 0, 1, 0, 1], candidates = [50, 60, 90] gives 60 (all 6 correct).