Two early pulses gave 3.3 percent higher mean relative normalized SEEG forecast error than five after contact order reversal in eight people at 20–500 ms

This study tested two versus five early calibration pulses on public STReEF SEEG data. It used eight people, 96 fixed conditions, and 6,803 bipolar readout-condition rows. Targets were five later pulses after recorded stimulation contact order reversal. An early-only scale reduced the mean waveform according to variation among its own calibration pulses. For the planned centered 20 to 500 ms score, two pulses gave 3.29% higher mean relative error than five. The descriptive 95% person-bootstrap interval was 2.42% to 4.26%. The exact boundary test supported a 5% mean-error allowance under symmetric-difference and independent-person assumptions (Holm p=0.0234). This used 60% fewer selected calibration inputs. A clear gain over zero was not shown (mean change -0.67%; interval -3.63% to +1.81%; Holm p=0.3594). The observed raw voltage error penalty was 11.89%; the later-mean target penalty was 12.25%. Also, 31 of 96 conditions exceeded a 5% penalty. Thus the allowance applies to the stated group score. The study reused a known cohort and a local plan; it had no external registration. Actual pulse omission, neural specificity, and BCI benefit were not tested. The package preserves code, all scores, three figures, compact exact products, source metadata, and pinned retrieval recipes. All 960 public voltage ranges were rebuilt in this run. Every input array and all 368 summary numbers matched exactly.

stereo-EEG · brain stimulation · forecast error · pulse calibration · brain-computer interface · OpenNeuro · noninferiority