Recall failures clustered in trials, while SEEG power tracked object recall count

We analyzed OpenNeuro ds006136 (OWM-Dataset, v1.0.1): 13 patients, 2134 unique trials, and 143 bipolar SEEG channels in middle temporal and middle frontal gyri. Each trial presented three objects. The variance-to-independence ratio for the number recalled was 1.21 (geometric mean), exceeded 1 in 12 of 13 patients (exact sign-flip p = 0.0005), and remained elevated under a block-shuffle drift control (p = 0.0027). Zero-recall trials occurred in 1.21% of trials, compared with 0.39% expected under independence (p = 0.012). Across all trials, encoding theta power (4–8 Hz) increased with recall count (mean rho = +0.065, Holm p = 0.022); 50–100 Hz power decreased during encoding (rho = −0.040, Holm p = 0.023) and early delay (rho = −0.041, Holm p = 0.002). After full-recall trials were removed, encoding theta and early-delay 50–100 Hz relations remained significant in 10 patients. Effect sizes were small (|rho| < 0.11), and channels within a patient were not treated as independent. The 50–100 Hz band includes line noise and may reflect aperiodic broadband change; no line-noise filtering or aperiodic fit was used. The data do not distinguish trial-level lapses from attention or changes in task difficulty.

working memory · object recall · SEEG · intracranial EEG · theta power · high-frequency power · trial-level lapses · OpenNeuro