A preprint reports that SENSE led EEG-to-speech benchmarks and showed EEG-text alignment, but reliable word-level reconstruction remains unresolved.