Spontaneous movements and their relationship to neural activity fluctuate with latent engagement states.

Switching between cognitive states is a natural tendency, even for trained experts. To test how cognitive states impact neural activity and behavior, we measured cortex-wide neural activity during decision-making in mice. During disengagement, neural activity was more variable across trials and could be better explained by a linear encoding model. This increase in explained variance during disengagement was associated with two changes: modestly stronger neural encoding of movements generally and an increase in task-independent movements specifically. Surprisingly, behavioral videos showed similar motion energy in both cognitive states. But while the overall amount of movements remained similar, movement alignment changed: as animals slipped into disengagement, their movements became less stereotyped. These idiosyncratic movements were a strong predictor of task performance and engagement. Taken together, our results suggest that the temporal structure of movement patterns constitutes an embodied signature of the cognitive state with a profound relationship to neural activity.