Fixed finite linear EEG features have sharp intervals under voltage deviation and adjacent sample change limits

We proved sharp intervals for fixed finite linear EEG features under voltage deviation and adjacent sample change limits. Holding later voltage at the deadline gives the center with the smallest worst possible error. A linear program gives the radius and traces that reach its endpoints. The tested centered filter had 1691 taps at 256 Hz. Limits were 100 microvolts and 1000 microvolts per second. At sample 25 after the event, the joint radius decreased by 15.30%, from 22.536068 to 19.088349 microvolts. The feature was a baseline-subtracted mean from 0 to less than 100 ms. Both limits held in 5.86% of trials, averaged equally across 13 adults. The 95% person bootstrap interval was 1.04% to 12.62%. Equal-person mean interval coverage was 99.67%. Coverage alone did not establish the assumptions. The audit used 24,071 events in 39 public runs from ds003061 version 1.1.2. All nine planned cases, code, lossless traces, source hashes, witnesses, and bootstrap indices are preserved. Independent checks passed. A complete repeat matched every numeric file. The method gives conditional measurement bounds. It does not estimate neural onset or validate limits for new trials.

electroencephalography · EEG · linear features · conditional bounds · minimax estimation · linear programming · signal processing · sample change limits · OpenNeuro