Two controlled AUC scores gave opposite SEEG reader rankings in two of 19 people before movement with eight contact pairs and penalty 10

We compared two choice AUC scores in public MetaRDK SEEG data. Each reader used eight bipolar contact pairs and penalty 10. One score averaged AUC within cue direction. The other averaged AUC within correct and error trials. We compared equal choice fit weights with equal cue and choice group fit weights. The readers and later test trials stayed fixed. The two scores gave opposite point rankings in two of 19 people before movement and two after movement. These counts are descriptive. They do not show which reader is better on new trials. We proved that AUC with equal test weight in all four cue and choice groups is the mean of the two controlled scores. The proof requires binary cue and choice, accuracy equal to their agreement, four nonempty groups, and half credit for ties. The identity passed independent checks for all 342 saved readers. We reproduced 190 overlapping fits from the earlier weight-rule study. The before-movement mean change within cue was +0.013462, with descriptive 95% person interval [-0.020016, +0.047829]. This test showed no clear reader gain. The package has code, compact data, source records, saved readers, predictions, four figures, and repeat instructions. All 14 numerical files replayed exactly. This exploratory offline study does not establish live signal timing or stimulation effects.

stereoelectroencephalography · SEEG · area under the curve · AUC · choice decoding · class weighting · cue balancing · neural signals