Despite several differences in contemporary learning theories and approaches to education, there is strong agreement that scaffolding and feedback are essential to foster successful learning processes. Technological advances have enabled the development of smart learning environments, e.g., intelligent tutoring systems, which provide authoring features for learning material with encapsulated assignments and customized automated guidance that directs learners to the current solution. In the field of STEM education, exploratory learning strategies are considered particularly effective. These approaches allow learners to construct their own knowledge—either individually or as a social group—by performing experiments or using engaging immersive 3D computer simulations. While basic simulations and virtual laboratories are widely accepted and used by physics teachers, most tools are very limited in terms of educational features such as internal guidance, feedback and assessment systems, whereas assignment sheets must be prepared separately and are not integrated with the simulation systems. Based on our work so far, we propose the flexible assessment system Antares (adaptive network-oriented tracking-based assessment for real-time educational simulations), which builds on a service-oriented architecture that targets real-time simulations. It allows for a flexible separation and reusability of several components of the learning system, such as the immersive 3D environment itself and the assessment engine. While a generalized, annotated immersive 3D environment can be reused for several similar experiments, the external assessment engine does not only offer enhanced, adaptive assessment measures—based on user actions and object states—but lets the teacher provide custom assignment sheets, which can be injected to compatible display elements within the immersive environment. First applications on a pendulum experiment demonstrate that the approach can be integrated with low expenditure, and the evaluation algorithm is able to match complex real-time measurement procedures.