Simulated fixed contact gain errors within five percent reduced SEEG site accuracy in five offline sessions

SEEG is stereoelectroencephalography. This study tested one digital gain factor per contact. Each factor stayed within five percent of one. One gain vector applied to all test pulses in a session. The unchanged reader identified known stimulation sites from 20–100 ms recorded responses. The five public sessions had 183 sites and 732 test pulses. Mean patient accuracy decreased from 40.6% to 29.5%. The saved vectors gave 77 fewer correct decisions in total. Testing only extreme gain settings missed a worse setting in three sessions. A mixed-integer search gave matching numerical minimum-accuracy bounds in four sessions. The fifth search reached its time limit. The remaining bound put minimum mean patient accuracy between 28.3% and 29.5%. Direct signal and distance checks agreed for 205,432 decisions. A new raw-data extraction matched all five saved signal archives exactly. The package preserves code, derived data, source hashes, gain vectors, predictions, solver states, figures and reproduction instructions. True test labels were used to choose adverse gains. This is a simulated offline calibration stress test on reused clinical data. It does not measure hardware drift or online interface performance.

stereoelectroencephalography · SEEG · contact gain · calibration · stimulation-site classification · mixed-integer optimization · offline stress test