Ten seconds of correlation calibration reduced mean Gaussian loss for transferred rest and film SEEG models in nine adults at 64 Hz

We tested transfer between paired rest and film SEEG records from nine adults in OpenNeuro ds003688, version 1.0.7. The fixed model mixed correlation from 50 seconds of source data with correlation from 10 seconds of early target data. The two matrices had equal weights. All models used the same early target mean and variance. The primary test used 275 bipolar signals with no shared contacts within an adult, at 64 Hz. The test block was 70 to 120 seconds after target onset. Mean Gaussian loss fell by 0.044810 nats per sample per signal. The 95% adult bootstrap interval was 0.007197 to 0.099045. Eight of nine adult means were positive. One adult supplied 53.90% of the net gain. Every mean after removal of one adult stayed positive. Ten fixed sensitivity checks are preserved. At a target weight of 0.75, the interval included zero. Later block normalization gave a smaller gain. This is an exploratory density score result in a reused clinical sample. It does not establish a neural cause, a film effect, or clinical benefit. The package preserves input data, source versions and hashes, code, fitted matrices, all scores, uncertainty calculations, figures, and reproduction instructions. A complete offline repeat reproduced all 20 result files exactly. A separate native reader reproduced four task arrays from two adults exactly, with one adult at each native rate.

stereoelectroencephalography · SEEG · covariance calibration · Gaussian density score · correlation transfer · OpenNeuro · clinical neuroscience