Refined tracking allows students and teachers to more accurately understand students' learning growth. To provide refined learning tracking with longitudinal diagnostic assessment, this article proposed a new model by incorporating probabilistic logic into longitudinal diagnostic modeling. Specifically, probabilistic attributes were used instead of binary attributes to model the latent variables that affect students' performance. Thus, in the proposed model, attribute-level growth can be quantified in a more refined manner. The feasibility of the proposed model was examined using simulated data. The results mainly indicated that the model parameters for the proposed model could be well recovered. An empirical example was conducted to illustrate the applicability and advantages of the proposed model. The results mainly indicated that when distinguishing the level of students, the diagnostic results of the proposed model and the conventional longitudinal diagnostic model for binary attributes displayed a high degree of consistency; however, the former could provide more refined description of growth and a better model-data fit than the latter.