Source interval checks accepted theta features from one-second windows with saved zeros in 250 selected SEEG records

We tested a fixed theta power feature in 250 selected SEEG records from seven people in OpenNeuro ds007095 version 1.0.0. Each feature used a one-second window. Feature ends were tested every 5 ms from zero to two seconds after the last marked stimulation interval. The source interval check accepted inputs that still contained saved zero runs. These inputs made up 17.70% of accepted windows, with equal person weight. A controlled test replaced samples in earlier SEEG with native zero codes. For each channel and record, we took the median error in affected windows. We then averaged channels, records, and people. The mean was 9.73%. A fixed threshold changed decisions in 0.93% of event-accepted windows. A mask that also rejected complete zero runs accepted fewer windows. Its accepted values were unchanged in the controlled test. A proof for a fixed squared norm shows when a flagged sample cannot affect the feature. The package preserves complete inputs, pinned sources, code, traces, three figures, and independent verification. All 12 numerical result files matched a clean repeat. The test does not establish a neural recovery time or decoder accuracy.

stereoelectroencephalography · SEEG · signal processing · theta power · feature validity · stimulation artifact · brain-computer interface