Theories of embodied cognition suggest that a shared environment and ongoing sensorimotor interaction are central for interpersonal learning and engagement. To investigate the embodied, distributed and hence dynamically unfolding nature of social cognitive capacities, we present a novel laboratory-based coordination task: the BallGame. Our paradigm requires continuous sensing and acting between two players who jointly steer a virtual ball around obstacles towards as many targets as possible. By analysing highly resolved measures of movement coordination and gaming behaviour, game-concurrent experience ratings, semi-structured interviews, and personality questionnaires, we reveal contributions from different levels of observation on social experience. In particular, successful coordination (number of targets collected) and intermittent periods of high versus low movement coordination (variability of relation) emerged as prominent predictors of social experience. Importantly, having the same (but incomplete) view on the game environment strengthened interpersonal coordination, whereas complementary views enhanced engagement and tended to generate more complex interactive behaviour. Overall, we find evidence for a critical balance between similarity and synchrony on the one hand, and variability and difference on the other, for successful engagement in social interactions. Finally, following participant reports, we highlight how interpersonal experience emerges from specific histories of coordination that are closely related to the interaction context in both space and time.