Full covariance did not establish a mean increase in SEEG response spread at 20 to 100 ms in five sessions under stable means and independent errors with zero mean

We compared response spread in two calibration noise metrics with public SEEG data from five patients. The main test used 138 remote stimulation sites, 15 bipolar recording pairs, and 1104 pulses. Four early pulses per site fitted the metrics. Four separate late pulses gave response moments at 20 to 100 ms. The full metric was C = 0.5 S + 0.5 D; the diagonal metric was D. The moment estimates are unbiased conditional on calibration under stable site means and independent late errors with expected value zero. The ratios can be biased. The mean full-minus-diagonal change in spread f was -0.0445. A descriptive interval from five patient point estimates was -0.1299 to +0.0299. The exact sign p value was 0.4375 under symmetric independent patient differences. The data did not establish a mean increase. They did not establish equivalence. In patient 5, spread fell while site accuracy increased by 9.375 percentage points. The late window and broader site set changed the sign of the mean spread effect. An invertible metric changes participation ratio while preserving response rank. Spread is thus a separate outcome from site reading and brain control. The package includes the conditional proof, voltage bins, source hashes and retrieval recipe, fitted states, predictions, tables, and three figures. Fresh raw extraction reproduced 30 input arrays and five event tables. A full replay reproduced 53 result files byte for byte. This is an exploratory extension on reused data. It does not establish live interface performance.

stereoelectroencephalography · SEEG · response spread · participation ratio · noise covariance · brain-computer interface · neural stimulation