This study explored the use of stealth assessments within a digital game to assess second language (L2) Chinese learners' reading comprehension. Log data tracking learners' in-game behaviors from a game designed for Chinese dual language immersion classrooms (Poole et al., 2022) were used to construct Bayesian Belief Networks to model reading comprehension. Variables included the use of an in-game glossary for word lookups, time spent reading texts, vocabulary knowledge, text length, response requirements, and the number of repetitions. Networks were first built using class-level gameplay data and subsequently applied to individual students. Results showed significant correlations between students' modeled comprehension and their performance on an external reading comprehension assessment (r = 0.52) as well as teachers' informal reading evaluations (r = 0.66), demonstrating stealth assessments' potential for unobtrusive reading comprehension measurement. The findings contribute to the understanding of digital game-based learning and assessment in language education, particularly in the context of reading comprehension for non-alphabetic languages like Mandarin Chinese.