Presented is a new approach to probabilistic modeling of a testing procedure, which allows considering the dynamics of subjects' abilities and difficulty of tasks during their solving. The modified Rasch function is used to build quantitative estimates, which arguments, unlike the classical version, are set in the probability scale. Methods of presented dependencies' identification using observational data are examined. As an illustration, the example of practical application of the proposed approach for modeling the Raven's Progressive Matrices testing procedure is given.