Early validation selected zero SEEG correction in seven of 15 additional MetaRDK records

We tested a fixed confidence forecast in 15 additional MetaRDK SEEG records. Early trials fit models and selected zero correction or a ridge correction from eight signals. The test used 1,502 later trials. Mean squared error was 0.0311343 for task inputs and 0.0308216 after selected SEEG correction. The mean gain was 0.0003127, or 1.00%. The descriptive 95% person bootstrap interval was -0.0000357 to 0.0007482; the exact sign-flip p value was 0.25. Three records improved, five worsened, and seven had zero correction. The early-mean constant forecast had less mean error than the selected forecast. Selection had less error than a forced penalty-10 correction in 13 of 15 records; this was a secondary check. All 537 main forecasts from the prior six records matched exactly. The package preserves two figures, all inputs and forecasts, model states, code, source versions, and retrieval steps. Fresh source checks rebuilt two records exactly. Source smoothing, normalization, sequential previous reports, and a future-click filter limit the result to this offline test. No stimulation benefit was tested.

SEEG · confidence forecasting · brain-computer interface · ridge regression · MetaRDK · forecast validation