In this study a data-driven probabilistic methodology is developed to assess the exceedance of mechanics-based alarm thresholds in full compliance with misclassification error probabilities required by the operator. Deck stiffness loss is adopted as the damage indicator, derived from accelerometer data collected from various sensors installed on the bridge. This indicator effectively detects, quantifies, and locates damage. Uncertainties in the collected data, stemming from loads, structural properties, and bridge response, are consistently addressed within a probabilistic framework. Physics-guided approach is used to define deck stiffness loss thresholds. These thresholds correspond to three alarm levels of increasing severity, with each exceedance subject to a predefined misclassification probability set by the operator. The framework is designed to meet stringent misclassification requirements by determining the appropriate sample size for each alarm level. The method demonstrates robust and consistent results for deck portions, the entire deck, and the entire bridge, as shown in numerical examples.