Early score costs reduced mean probability change for eight fixed SEEG readers at a 32-bit sample budget

We tested voltage coding for eight fixed SEEG state readers. The public data were OpenNeuro ds004473, version 1.0.2. Early trials set fixed pair widths from squared pair score errors. Later trials tested the change from each native reader probability. All three rules used 32 payload bits per simultaneous eight-pair sample. The score rule reduced equal-patient mean absolute probability change from 0.09905 to 0.06609. The paired difference was -0.03296, with a descriptive 95% patient bootstrap interval of -0.05253 to -0.01649. Seven recordings improved. One received the equal allocation. Every primary packet contained 2,056 bytes. The voltage cost rule gave smaller waveform error but larger mean probability change. Task accuracy did not show a clear improvement. The package preserves digital buffers, early costs, allocations, 4,044 packets, predictions, three figures, source hashes, and all code. A complete offline rerun matched 49 computed files exactly. This is an exploratory reanalysis of reused trials from one recording per patient. Live control, hardware cost, and stimulation were not tested.

stereo-EEG · signal compression · brain-computer interfaces · quantization · probability fidelity · fixed decoder · mixed-width allocation