Source identity checks removed exact mean bias from uniform bootstrap draws of residual EEG covariance in 19 ds004408 listeners

We tested a uniform listener bootstrap for mean residual EEG covariance. The fixed matrix used 19 listeners from OpenNeuro ds004408, version 1.0.8. It used two fixed 150-second speech excerpts, 56 channels, and a 1–4 Hz filter. A fixed speech envelope model was removed from each standardized signal. The distinct-listener mean was 0.005723106. Repeated copies of a listener raised the exact expected mean to 0.057485504. Removal of numeric unit scores did not change this result because residual self scores were below one. Removal of pairs with the same source identity preserved the distinct-listener mean. This result is conditional on a draw with at least two distinct sources. A proof applies to any fixed finite symmetric matrix and uniform independent draws with replacement. The package provides exact moments, a source identity tool, 50,000 saved draws, processed input data, source hashes, and a raw data retrieval recipe. Independent checks and a full offline repeat passed. Two fresh raw files reproduced the saved input arrays exactly. The study does not establish confidence interval coverage, a biological response, or absolute EEG amplitude calibration. Repeated-person pairs are a known issue. This work supplies a residual covariance test and a reusable audit method.

bootstrap · source identity · residual covariance · EEG · intersubject correlation · resampling · exact expectation · OpenNeuro ds004408