Before-pulse covariance gave both higher and lower SEEG site accuracy than response covariance with two fit pulses per site in five offline sessions

Before-pulse covariance gave mixed site accuracy in five public SEEG sessions from OpenNeuro ds004457, version 1.0.2. We used fixed contacts, 183 known stimulation sites, and 732 test pulses. Each site had two fit pulses and four later test pulses. The response window was 20 to 100 ms. Both main matrices used 50% diagonal shrinkage and the same response templates. Equal-patient accuracy was 33.44% with before-pulse covariance and 35.00% with response covariance. The difference was -1.55 percentage points. Patient differences ranged from -13.79 to +4.29 points. The descriptive patient bootstrap interval was -7.99 to +3.35 points. This result does not establish equal performance or a common benefit. A secondary one-pulse check gave 25.95% accuracy with before-pulse covariance and 23.28% with equal feature weights. The allocation was offline, and background site associations remain unresolved. Fresh raw extraction matched the cited inputs. Independent distance checks and a full analysis replay passed. The package preserves code, voltage inputs, source hashes, trial scores, fitted readers, and three figures.

SEEG · electrical stimulation · covariance · site classification · intracranial EEG · offline evaluation