Games hold promise as assessment tools, but the mechanics used must first be validated. Accordingly, we examined a novel implementation of an unbounded number line estimation task to assess students’ fraction comparison competence, laying the groundwork for designing a stealth assessment for math games. Sixty-eight Finnish 6th graders played a math game in which they estimated fraction magnitudes on an unbounded number line. Additionally, the students completed a non-game assessment based on a traditional fraction comparison task using the same fractions, as a part of a counterbalanced design. Overall, the results showed many similarities between the assessment conditions, including a typically observed distance effect, a similar correlation with general math achievement, and a moderate positive correlation between the conditions. However, comparison performance was lower in the game assessment. Further, the performance classifications based on whole number ordering consistent and inconsistent comparison tasks showed only a slight agreement between the game and non-game assessments. This suggests that the two approaches may capture different aspects of performance, with the game assessment potentially yielding misleading learning analytics and inappropriate pedagogical decisions. Also, motivational measures suggested challenges with the unbounded tasks, as students’ flow experience was significantly lower in the game condition. The results underscore the need to carefully evaluate theoretically informed assessment mechanics and highlight the importance of further research to establish the unbounded number line task as a valid measure of fraction comparison competence.