Exact mean correction had greater power error than interpolation under simulated adaptive gaps in two SEEG recordings

Known positive probabilities gave zero expected bias for finite-window SEEG power under the specified simulated gap rule. The test used two public rest recordings and 16 contact pairs per person. Each packet kept an anchor. It removed four other samples with a probability set by that anchor. The main band was 70 to below 140 Hz. Bins from 98 to below 102 Hz were excluded. The test used 128 masks per pair and test second. Logged correction had greater root mean squared error than linear interpolation in every selected pair. For one-second outputs, error divided by complete pair mean power was 142.86 and 172.42 for correction. The interpolation ratios were 0.237 and 0.311. Ten-second averaging reduced correction error, but interpolation still gave lower error. The package includes the proof, derived data, retrieval code, tables, four figures, and independent checks. A separate replay reproduced all numerical results exactly. The test used simulated gaps in rest data. It did not test real stimulation, task decoding, or patient benefit.

SEEG · signal-dependent gaps · power spectral density · inverse inclusion weighting · brain-computer interface · interpolation · missing samples