Adding a band power slope constraint raised mean Brier error in eight offline SEEG tests with fixed early training

We tested three small readers on eight public SEEG records from OpenNeuro ds004473 version 1.0.2. The readers used the same eight pairs, early training trials, feature scales, and coefficient penalty. Adding a band power slope constraint to an offset constraint raised mean Brier error from 0.2111 to 0.2486. The paired increase was 0.0375, with a descriptive 95% person bootstrap interval of 0.0081 to 0.0768. Mean balanced accuracy fell from 67.0% to 56.8%. The added constraint canceled changes under a stated three-band offset and slope model. It did not cancel power transfer before band integration. A spectrum check changed output probabilities by up to 5.27 percentage points in stored raw audit windows. All 64 random controls of the same reduced dimension had lower mean probability error. These controls were added after the primary result. The package preserves code, source pins, features, model states, predictions, proofs, four figures, and reproduction instructions. The evidence is limited to reused offline records with response-aligned labels and fixed early training.

stereoelectroencephalography · brain-computer interface · band power · logistic regression · Brier error · signal robustness