Mean EEG and fMRI signals did not pass a shared class transfer test after cubic log RT adjustment in 25 and 10 reserved adults

We tested stable stimulus class information in coarse mean brain signals. Separate public adult studies supplied scalp EEG and BOLD fMRI during flanker tasks. Development used eight EEG people and six MRI people. A final local code lock preceded reserved neural extraction. The fixed rules included 25 of 32 reserved EEG people and 10 of 20 reserved MRI people. A reader learned the class-effect direction in one task period and tested the other. Cubic log response time and trial order were adjusted in each period. Mean Fisher z scores were +0.009081 and +0.026596; pointwise 95% person Student limits were [-0.028792, +0.046953] and [-0.119946, +0.173139]. The shared test had Holm p=0.385516. The fixed shared rule failed. No positive generalization across the two modalities was established. An MRI source-field comparison had Holm p=0.385516. Both recorded label fields, all negative people and failed checks remain. Development included a corrected event-map error and a restriction to means after a rich MRI reader failed a local signal check. A defined model gave test correlation 0.496 after RT adjustment although its neural variable was independent of class before adjustment. This is an interpretation limit, not a fitted human mechanism. Separate cohorts, source labels, artifacts, short MRI runs and measurement models limit scope. Code, aggregate results, permitted MRI derivatives, source pins and repeat instructions are supplied. EEG signals and per-trial neural features remain external under the source reuse notices.

human neuroscience · EEG · fMRI · flanker task · response time · generalization · replication · conditional decoding