When constructing game-based virtual learning environments, developers face considerations that differ from the concerns instructional designers raise for traditional learning scenarios. Absent are the physical world cues that teachers use to monitor guided learner actions. Without the sights and sounds of the real world, game developers rely on machine-aggregated server data and as such on impersonal feedback to interpret player actions and conclude latent learning activities. Embedding formative methods as vivid element of assessing and revising learning outcomes is considered as value added approach to enhance traditional learning analytics. Providing a holistic view of behavior in learning games by fostering the understanding of player activity also infers to instructional design issues. Merging insights from both perceptions opens the way to sustainable optimize learning games design. The study exemplifies the integration of data-driven methods by Zon, an MMORPG to learn Mandarin.