SEEG word-encoding patterns drifted beyond serial-position effects, but drift did not predict recall contiguity

We tested whether neural context drift during word encoding in delayed free recall is detectable in SEEG after controlling for serial position, and whether drift predicts temporal clustering in recall. We used session 0 from OpenNeuro ds004789 (RAM FR1), reading acq-bipolar EDF files with HTTP byte ranges. Of 154 eligible participants, 44 were complete at a fixed stop time; 38 had at least 10 recall transitions. For each word, we measured log10 power in eight frequency bands from 0–1.6 s after onset in the first 32 SEEG bipolar pairs. Within-list similarity declined faster with lag than similarity across lists matched for serial-position difference (mean slope difference D = −0.00457, 95% CI [−0.00566, −0.00354]; 41/44 negative; Holm p < 3×10⁻⁵). Adjacent within-list similarity exceeded the control (0.0819, 95% CI [0.0694, 0.0962]; 44/44 positive; Holm p < 3×10⁻⁵). Drift magnitude did not predict temporal factor score (Spearman rho = 0.028, n = 38, p = 0.43). The control addresses serial-position effects but not elapsed time: within-list pairs at lag L were about 2.5×L seconds apart, while across-list pairs were minutes apart. Slow signal non-stationarity could therefore produce the observed drift. The analysis used one session, the first 32 SEEG pairs without anatomical selection, and the first completed participants in ID order; the sample was not random.

SEEG · intracranial EEG · neural context drift · free recall · temporal clustering · serial position · RAM FR1 · OpenNeuro