Proprioceptive cortical neurons implement optimal state estimation.

While the primary somatosensory cortex (S1) is known to be essential for skilled limb movements, it remains unclear how distinct functional ensembles within its network uniquely contribute to motor control. By isolating a stable population of proprioceptive S1 neurons in layer 2/3, we demonstrate how their selective removal impacts goal-directed reaching. We found that following their microablation, global reach kinematics remained stable, yet natural movement variability was reorganized. Trajectories became more spatially dispersed while simultaneously becoming more geometrically stereotyped. Through computational modeling, we show that this seemingly paradoxical increase in one form of variability alongside a reduction in another reflects a failure in optimal state estimation. In contrast, broad S1 lesions mask the unique contribution of this ensemble, revealing instead that cortical feedback is temporally phased, engaging during high-precision grasping and retrieval rather than initial reaching. These findings identify a functionally defined subset of cortical neurons as the biological substrate for a core postulate of optimal control theory, operating within a broader S1 circuit specialized for fine motor refinement.