Responses 1 and 4 gave 3.4% higher mean SEEG forecast error than responses 2 and 3 in five sessions at 20–500 ms

We tested which two recorded responses to use for stimulation-response calibration. We used OpenNeuro ds004457 version 1.0.2. At 138 sites in five patients, the first four accepted responses supplied calibration. Responses 1 and 4 formed a wider selection. Responses 2 and 3 formed an adjacent selection. Both estimates used two responses and were scored on 884 later responses at the same site and current. The main window was 20–500 ms. Each site error was scaled by its four-response calibration energy. The equal-patient mean error reduction from the wider selection was -3.44%, with a 95% patient bootstrap range of -9.02% to +3.23%. An exact mean-time match gave -4.05%. Recording pairs with no shared contacts gave +1.96%. These results do not establish a general spacing benefit or cost. The four-response reference had lower error than the adjacent estimate in all five patients. We verified 110 raw source files and rebuilt all input arrays and event tables exactly. Independent scalar scores and a full repeat analysis passed. The package contains a risk proof, three figures, voltage bins, forecasts and targets, tables, code, source hashes, and reproduction instructions. This reused-cohort study selects measurements from existing pulses. It does not test a change in delivered stimulation timing.

stimulation-response calibration · stereo-EEG · waveform forecast error · measurement selection · ventral temporal cortex