Scores reduced mean Brier error with eight adaptation labels in eight offline SEEG tests at a simulated 10% task rate

We tested task rate estimates with public SEEG data from eight people. A fixed reader used early labeled trial blocks. Eight random labels were then checked in a later block. We compared estimates from these labels alone with estimates that also used the remaining window scores. The test used a simulated 10% pre-response rate and a final trial block. Mean Brier error fell from 0.0831 to 0.0811. Adding scores helped in 6 of eight recordings. The paired mean difference was -0.00206. The bootstrap 95% interval was -0.00349 to -0.00053. The exact sign-flip test gave p = 0.055. These two uncertainty measures give mixed evidence. The result does not establish a population benefit. The package contains code, features, source records, simulations, and three figures. All 2,014 retained trial feature arrays were rebuilt from public signal ranges with zero difference. The test reused a public cohort. It did not test continuous device control or brain stimulation.

label shift · Bayesian prevalence estimation · Brier score · intracranial EEG · SEEG · brain-computer interface · audit labels · simulation