Early pulse projection reduced 20–100 ms site accuracy in two of four offline SEEG recordings

A filter removed one spatial direction fitted to 0–10 ms calibration voltage. The test used public OpenNeuro ds004457, version 1.0.2. Four recordings had 98 remote 6 mA sites and 392 withheld pulses. Each site reader was fitted again after removal. Later site accuracy decreased in two recordings and increased in two. The equal-recording difference from 512 matched random removals per recording was −8.68 percentage points. The descriptive 95% interval was −20.76 to +3.41 points. Early test voltage energy removal averaged 66.18%. The four-recording sample does not establish a population effect. Site information can come from artifact, neural response, or drift. The package includes source code, input bins, predictions, random controls, figures, source hashes, and a successful raw data rebuild. The method tests early energy removal and later information preservation as separate endpoints.

stimulation artifact · SEEG · spatial projection · signal processing · site decoding · matched controls