Five-pulse test averaging changed mean SEEG skill signs in three of eight people under five-pulse calibration and equal condition weights at 20–500 ms

We tested how pulse averaging changes stimulation-response forecast scores in public SEEG data. The fixed STReEF sample contained eight people, 96 conditions, and 960 pulses. Each five-pulse forecast predicted targets from the other recorded contact order. The main window was 20 to 500 ms. Conditions and people had equal weights. Mean normalized error was 0.9973 for single-pulse targets and 1.3213 for five-pulse mean targets. Each target used its own zero-waveform denominator. Mean skill changed from positive to negative in three people. No condition changed its skill sign. A finite-subset identity shows that the absolute gain over zero cannot change with target averaging. The person sign changes arose from changes in condition score scales. The package supplies the proof, a fixed-denominator check, all calibration and target sizes from one to five, and three figures. Full checks rebuilt all 96 conditions from 960 pinned public voltage ranges. All arrays and range hashes matched. A separate analysis replay matched all seven numerical files. The sample was reused. Contact order and time changed together. The result applies to the specified offline test.

stereoelectroencephalography · electrical stimulation · pulse averaging · forecast evaluation · normalized error · prediction skill · finite-sample audit · brain-computer interfaces