A fixed SEEG reader had mean recall AUC of 0.547 for random words and 0.544 for complete lists in eight sessions

We tested whether the test unit changes the measured accuracy of a fixed SEEG memory reader. We used 1,920 words from eight session-zero recordings in public OpenNeuro ds004809 version 2.2.0. Each reader used 16 bipolar SEEG pairs and six power bands. Random-word and complete-list tests both trained on 192 words per fold. Across ten repeats, mean recall AUC was 0.547 and 0.544. The mean difference was 0.003. Its descriptive participant bootstrap 95 percent interval was -0.011 to 0.017. The sample does not establish a mean split effect or equivalence. A position-only reader had mean AUC of 0.587 and 0.583. A list-identity control gave tied scores within every complete-list test fold. Its AUC was 0.500 within folds and 0.435 when scores from different folds were pooled. A proof explains this difference under the fixed linear reader assumptions. The package preserves inputs, source versions and hashes, all fitted models, controls, three figures, and code. An independent EDF reader rebuilt all features for one participant. A separate analysis run reproduced the numerical outputs. These are offline tests within a session. They do not establish online BCI performance or a stimulation effect.

stereo-EEG · memory · free recall · AUC · cross-validation · brain-computer interface · score pooling