Three pyFR SEEG sessions showed no frozen FR1 recall transfer; spectral tilt did not predict recall within serial position

This planned transfer analysis tested whether spectral subsequent-memory read-outs defined on ds004789 FR1 data from 18 patients predicted free recall in ds004865 pyFR sessions without fitting to the new patients. Because S3 bandwidth was limited, the fetch processed sessions in order of read cost and stopped at a 780-second limit; only three sessions were analyzed, so they are not a random sample of the 20 SEEG subjects. The frozen FR1 logistic read-out had mean AUC 0.513 (2/3 sessions above 0.5; exact one-sided Wilcoxon p = 0.25) and did not provide evidence of transfer in these sessions. The untrained high-frequency minus theta spectral tilt had mean AUC 0.562 (3/3 above 0.5; Wilcoxon p = 0.125, the smallest possible p for N = 3). A combined within-list permutation test added after the plan gave p = 0.012, but the result came mostly from one session in which tilt declined with serial position; its within-serial-position AUC was 0.500, and the combined serial-position-controlled permutation test gave p = 0.26. The tilt result therefore cannot be separated from a primacy signal in these data. Two sessions had sampling rates below 1000 Hz, the analysis used shorter mirror-padded read windows than the FR1 pipeline, and this three-session result does not test transfer to the full ds004865 cohort. The data were fetched from the live OpenNeuro S3 path; the exact dataset release is uncertain. Per-file URLs, ETags, hashes, and byte ranges are retained with the features.

free recall · SEEG · subsequent memory · cross-dataset transfer · spectral power · serial position · OpenNeuro