A contact metric kept geometric SEEG templates and site labels fixed across full layouts of consistent contact differences in four records

We tested whether a software pair layout changes whole-vector geometric median templates from the same public SEEG contact signals. The primary test used 134 stimulation sites and 536 test pulses from four records of OpenNeuro ds004457 version 1.0.2. Each layout retained the same full contact contrast space. In the 20 to 100 ms window, a star layout changed 46 of 536 predicted site labels compared with adjacent pairs. A complete layout changed 25. A contact metric kept every template and label fixed to numerical precision. A fifth record with identical layouts gave a negative control. The reader used one fixed contact covariance. Equal-record accuracy was 42.96% for the contact median and 44.21% for the arithmetic mean. These tests did not establish an accuracy gain. The proof requires consistent pair signals and the same full contact contrast space. New raw downloads passed all source hashes. Raw extraction and a fresh analysis reproduced all 820 result arrays exactly. Independent numerical checks also passed. The package preserves compact data, source hashes, retrieval code, fitted templates, site decisions, the proof, and three figures. These reused clinical records do not give a prospective BCI performance estimate.

stereoelectroencephalography · geometric median · bipolar recordings · contact metric · response templates · site classification