Shared scores gave no clear Brier error gain for 100–500 ms keypress forecasts in eight SEEG recordings with known cues

We tested one shared logistic score against five separate, order-corrected scores in eight public SEEG recordings. Known cues and past responses were inputs. SEEG powers supplied 32 more inputs. For 100, 200, 300, 400, and 500 ms keypress forecasts, shared scores used 66 coefficient slots instead of 310. The test had 339 later trials and 2,371 endpoints. People had equal weight. Mean joint Brier error was 0.120488 for the corrected reference and 0.122502 for shared scores. The change was +0.002013. A descriptive 95 percent person bootstrap interval was [-0.002093, +0.007020]. Four of eight people improved. This result does not show a clear gain or equal accuracy. The package includes code, inputs, forecasts, figures, a probability order proof, and source checks. A full analysis repeat matched all forecast arrays and ten numerical files exactly. This fixed task test concerns brain reading. It does not test live control, rest commands, or brain stimulation.

stereoelectroencephalography · keypress forecasting · Brier score · logistic regression · brain-computer interface · probability calibration · response history