A fixed power cap raised mean later 70–110 Hz reconstruction scores in seven SEEG rest records with separate input contacts

We tested a fixed power cap for later SEEG voltage prediction in seven rest records from OpenNeuro ds006065 version 1.0.0. We kept 511 target pairs. Two input pairs shared no target contact. Early data set the fit and cap. Later target data did not enter the cap. For 70–110 Hz, equal person mean R² rose from −1.212 to 0.005. All seven person means increased. The paired gain was 1.217. Its descriptive 95% person bootstrap interval was 0.074 to 2.735. The cap changed 4.45% of test blocks. Five fixed scores were below −10; no capped score was below −10. Six target scores worsened. The final mean interval spanned zero. The result supports error containment in these records. It does not establish useful signal recovery or a stimulation benefit. A proof gives the power bound and shows why error can increase. The package preserves code, complete block results, figures, source hashes, and rebuild instructions. All seven original voltage files were checked again. A full rebuild reproduced all results. This is an exploratory extension in the same group of people as an earlier benchmark.

stereoelectroencephalography · SEEG · signal reconstruction · power cap · brain-computer interface · R²