We propose an adaptation of the Glicko-2 rating system in a K-12 math learning software setting, where variable time intervals between solution attempts and the stratification of student-item pairings by grade levels necessitate modification of the original model. The discrete-time stochastic process underlying the original system has been modified into a continuous-time process to account for the irregularity of intervals between solution attempts. Also, conceptual prerequisite relationships between items were used to provide initial rating estimates that allow for rating values to be meaningfully compared across grade levels. Fitting the model using real student learning data results in rating value distributions successfully exhibiting a gradation with the increase of grade level. A potential area of application in a personalized education setting is also briefly discussed. [For the full proceedings, see ED615472.]