Pre-Model Structure Screening for Non-Stationary EEG Time Series

Electroencephalography (EEG) time series are highly non-stationary and often analyzed using models that implicitly assume stable underlying structure. This work introduces a pre-model structure screening approach for EEG data, designed to assess whether and when such modeling assumptions may be justified prior to interpretation or prediction. Using a magnitude-based coherence index applied in a sliding-window framework, the method evaluates temporal organization in raw EEG signals without relying on frequency bands, task labels, decoding objectives, or predictive modeling. Applied to a publicly available motor movement/imagery EEG dataset, the analysis demonstrates that, despite the visually noisy and irregular appearance of raw EEG waveforms, structured temporal regimes can be detected through phase-agnostic transformations of the signal. The contribution of this work is methodological rather than inferential. The proposed coherence index is intended as an upstream diagnostic tool that quantifies time-varying structural validity in non-stationary EEG recordings, informing when downstream models may be more or less appropriate. More broadly, this framework reframes coherence detection as a gatekeeping step in EEG analysis pipelines, emphasizing structural assessment prior to modeling rather than post-hoc interpretation.