Observability constrains expertise-dependent kinematic readout in action prediction

Human movement is variable even when the latent intention is identical, which complicates action prediction from kinematics. Although perceptual expertise often improves anticipation, it remains unclear how this benefit is modulated by intrinsic trial-to-trial fluctuations in kinematic structure. We introduce kinematic observability, operationalized as the distance of a movement instance from the optimal decision boundary in kinematic feature space. In a table-tennis action prediction task with an expert-novice paradigm, we sampled trials based on objective evidence strength rather than masking body parts. Experts outperformed novices on high-observability trials and showed readout weights more closely aligned with the optimal encoding direction. However, under low observability, the expert advantage was attenuated, accompanied by a criterion shift reflecting increased reliance on prior expectations. These results indicate that expertise improves anticipatory readout when observability is high but yields smaller benefits when trial-wise kinematic evidence is weak; under low observability, experts also shift decision criterion in a manner consistent with increased reliance on prior expectations.