Two past seconds increased mean unsent SEEG power error by 5.8 percent for linear readers in eight people with eight sent streams and 120 calibration seconds

We tested whether two past seconds help estimate unsent SEEG log power. The source was OpenNeuro ds005624 version 1.0.0. Eight people each had two sessions. A linear reader sent eight power streams. Its input bank used old calibration. Each new model had 120 calibration seconds. An early time block selected the ridge penalty. Later separate clips supplied 240 scored seconds per person. No history crossed a clip gap. The reader with current values and two past seconds had 5.8 percent higher mean error than the reader with current values only. Mean normalized squared error was 1.269341 with history and 1.200023 without history. Error increased in seven of eight people. The mean paired error increase was 0.069318. Its descriptive 95 percent person bootstrap interval was 0.032870 to 0.109649. The package preserves code, all models and predictions, tables, three figures and pinned source retrieval records. Fresh retrieval checked all 576 source ranges. All rebuilt feature and witness arrays matched the stored data exactly. These recordings have prior use. The result is exploratory. It applies to the tested power readers. It does not establish intention decoding, live control or a stimulation effect.

SEEG · signal reconstruction · power estimation · ridge regression · time history · OpenNeuro