Both runs increase the cosine between contrast weights in all 26 tested ds000102 models.

This event-design analysis compares contrast weights derived from the `trial_type` and `Stimulus` condition fields in OpenNeuro ds000102. It uses all 26 participants and both runs, with 1,248 trials, and no measured imaging signal. The fields agree on all 624 run-1 trials and conflict on 304 of 624 run-2 trials. In the primary ordinary least-squares model, the median cosine between field-specific weights is 0.055053150673 for run 2 and 0.525788869078 for a model with both runs. The cosine is larger with both runs for all 26 participants; nine run-2 cosines are negative and all two-run cosines are positive. The two-run weights still differ, and run 1 adds an identical component under both fields, so the increase does not identify the correct field. All eight planned settings retain the same negative-cosine counts and direction of increase. These are deterministic summaries of this complete release, not population confidence intervals. The models omit measured signal, motion terms, censoring, temporal whitening, HRF derivatives, and measured noise covariance, and assume a common contrast coefficient across runs. The displayed run-2 condition remains unresolved.

functional MRI · contrast weights · event design · condition labels · ordinary least squares · OpenNeuro · ds000102 · model sensitivity