The two-run model increases the contrast weight cosine for all 26 ds000102 participants in 17 settings.

This reproduction compares contrast-weight alignment for two event-condition fields in the OpenNeuro/OpenfMRI ds000102 flanker task. The fields agree on all 624 run-1 events and disagree on 304 of 624 run-2 events. In the primary setting, the median run-2 cosine is 0.055053150672911624 (range -0.4942117404597276 to 0.4873386948068449; 9/26 negative). The two-run median is 0.525788869078214 (range 0.2359437605774073 to 0.7489194136592704), and the cosine increases for all 26 participants in each of 17 settings. A new implementation reproduces 702 model rows and 7,592 primary frame rows; all 38,090 numerical and rank comparisons pass, with maximum numerical metric difference 7.105427357601002e-15. Analytic integration preserves cosine signs in 624 additional model rows. These deterministic design calculations use no measured brain responses, do not establish which condition labels are correct, and have no sampling confidence intervals. The result applies to the fixed sample and stated model assumptions; it does not establish biological causality or reproduce neural findings.

flanker task · contrast weights · event design · cosine similarity · fMRI · reproducibility · condition labels · OpenNeuro