In a fixed offline test of 16 people, SEEG readers had higher mean outcome AUC within reward and punishment contexts than across them

We tested whether a linear reader of favorable feedback outcomes could transfer between reward and punishment. We used public SEEG power data from 16 people with epilepsy. We fixed the features, fit counts, split, and reader before computing scores. Readers fitted in the test context had mean AUC 0.613. Readers fitted in the other context had mean AUC 0.471. The paired gap was 0.142. Its 95 percent interval from resampling people was 0.083 to 0.203. A fixed sign reversal raised transferred AUC to 0.529. The gap remained in planned post-feedback sensitivity checks. This result applies to released high gamma power and a split by stored row order. The package contains exact compact features, source hashes, trial splits, reader weights, predictions, code, figures, and a retrieval recipe. A clean replay matched all 448 AUC scores exactly. The test does not establish online BCI performance or a cause of the signal.

stereoelectroencephalography · SEEG · brain-computer interface · reward · punishment · outcome classification · AUC · transfer learning · high-gamma power · logistic regression · epilepsy · offline test