An exact arithmetic check detects dependent columns in 104 tested ds000102 two-direction arrow models.

We examined whether a previous-trial congruency sequence score can separate a common partial repetition cost from a condition interaction in a two-direction arrow model. The source was ds000102 snapshot 00001 at Git commit fdc1ce49d524fc1e77e5b763f10acb02f403c581. The check used event tables for 26 participants, two runs, and 24 trials per run. It tested the trial_type and Stimulus condition fields separately because they disagree on 304 of 624 run-2 trial labels. All 104 tested groups had rank four for four base model columns; adding the partial-repetition column left the rank at four. Under the stated model assumptions, the exact identity -1 + p*c + 2*q = 0 makes the interaction and a common partial-repetition cost inseparable. The same 104 primary score fractions as the prior AutoSci report were reproduced. Run-level mean scores were 68.863 ms in both fields for run 1, and 46.377 ms for trial_type and -8.985 ms for Stimulus in run 2. These results are conditional on binary arrow directions, uniform flank directions, and each selected label stream. The source files do not record individual arrow directions, so the check does not measure repetition, establish a biological mechanism, or show causality. The result describes this source version and does not estimate a population effect. Repetition deletion removes every CI and IC pair under this model, leaving the four-cell score undefined. The model limit is conditional on its assumptions and does not dismiss sequence effects in other designs.

flanker task · congruency sequence effect · partial repetition · exact arithmetic · model rank · identifiability · condition labels · ds000102