The CPz N400 mean stayed below zero after any allowed removal of 207 or fewer trials in 31 ERP CORE adults, with a 300–500 ms score, P9/P10 reference, and at least ten trials per person and condition

We tested further removal of trials from a fixed N400 voltage score in public ERP CORE data. The primary screen retained 2,079 trials from 31 adults. The score used CPz, 300–500 ms, and a P9/P10 reference. People had equal weight. Each person had to keep at least ten trials in each condition. The observed unrelated minus related mean was −7.867112 µV. For any allowed removal of 207 or fewer trials, the exact limits were −14.249632 and −1.289131 µV. The first possible reversal needed 267 removals, or 12.842713 percent. All eight specified sensitivity checks also had negative upper limits at their own 10 percent budgets. The method used exact integer arithmetic and supplied trial subsets that reached the limits. Checks included exhaustive small cases, a separate calculation, a complete repeat, and fresh reconstruction of all 4,800 epochs. These limits apply to the saved scores and fixed people. They do not establish artifact-free signals or a population effect. The package preserves code, a proof, scores, complete limits, source hashes, figures, and reproduction instructions.

N400 · event-related potentials · EEG · trial removal · sensitivity analysis · exact bounds · dynamic programming · ERP CORE