A site blend fitted with five pulses did not establish lower 20–500 ms SEEG forecast error in eight sessions

We tested whether recording pairs at one stimulation site can share scales fitted from five early pulses. We used public SEEG from eight epilepsy patient sessions in OpenNeuro ds004696, version 1.0.1. The study covered 86 sites and 886 stimulation-to-recording pairs. Each test used the final five good pulses. In the 20–500 ms window, the selected blend reduced mean patient error by 0.271 percent from separate pair scaling. Five patients improved, two worsened, and one had no change. The two-sided patient sign-flip test gave p = 0.0625, under independent-patient and sign-symmetry assumptions. This result does not establish lower error across patients. Full sharing increased mean patient error by 0.930 percent. The package supplies a site scale tool, a risk identity under a stable response model, code, exact input products, all individual scores, figures, source hashes, and retrieval instructions. Fresh source checks for patients 01 and 08 matched every saved array and event map. Other inputs matched their parent archive hashes. The later pulses had been used in prior work. This is an exploratory comparison. No live brain-computer interface or clinical outcome was tested.

SEEG · voltage forecast · calibration · scale sharing · brain-computer interface · pulse response · signal processing · exploratory study