Image code matching at P7 and P8 had 1.56 percent accuracy across two ERP CORE groups

We tested image code agreement across two separate groups in the public ERP CORE N170 task. The fixed analysis used P7 and P8 from 100 to less than 250 ms. It averaged two accepted occurrences and removed category means. The sample had 37 people and 5,415 accepted code pairs. No person was in both groups. Both groups contained all 160 codes. The image agreement correlation was 0.02898. Its approximate 95 percent person-bootstrap interval was -0.03601 to 0.07383. The one-sided permutation p value was 0.2077. Nearest matching gave five exact hits in 320 comparisons, or 1.5625 percent. Uniform code chance was 2.5 percent. This test did not show useful image matching. It does not prove that the records have no image information. For a fixed nearest-code rule, independent fair binary labels have exact expected accuracy 0.5 + h/2, where h is identity accuracy. The package includes the proof, a split overlap tool, epochs, source hashes, code, results, and repeat instructions. All 12 numerical files had equal bytes in a full repeat. Fresh downloads reproduced two source records exactly.

EEG · ERP CORE · N170 · event-related potentials · shared stimuli · image matching · cross-participant agreement · permutation test · person bootstrap · chance accuracy · data leakage · split audit