Decoding Motor Action Timing and Metacognitive Signals from Single-Trial EEG Using a Transformer-Based Model (EEG-ViT)
Metacognitive self-evaluation, the capacity to assess ones own performance without external feedback, is fundamental to adaptive cognition. A central unresolved question in metacognition is whether self-evaluation reflects a direct readout of the neural signals supporting first-order performance, or whether it depends on partially distinct higher-order representations. Unusual approach of monitoring actions in a continuous dimension, such as time, is particularly useful because errors are graded rather than binary, allowing metacognition to be treated as a quantitative estimate of error magnitude. Temporal error monitoring (TEM), in which individuals judge the accuracy of self-generated time intervals using only internal signals, provides a tractable framework for isolating metacognitive evaluation from external correction. We asked whether TEM engages the same neural mechanisms as first-order timing performance, the execution and control of the timed action itself, or recruits at least partially distinct processes involving additional evaluative computation. We decode single-trial EEG using a Vision Transformer applied to PCA-optimized {theta}, , and {beta} signals recorded during both monitoring of motor timing. First-order timing was decodable from each frequency band independently, whereas TEM required simultaneous integration across all three bands. Individuals for whom TEM neural states were decoded with greater accuracy demonstrated stronger behavioral error-monitoring precision. These findings establish TEM as a neurally distinct re-representation of timing performance, demonstrating that metacognitive self-evaluation requires evaluative computations beyond those supporting motor action execution.