Two methods agree on cue predictor changes to 1e-9 in 126 ds000030 control runs.

A replay of archived code and an independent numerical implementation were compared for stop-signal cue predictors in 126 control runs from OpenNeuro ds000030 version 1.0.0. The analysis uses public task tables, fixed trial outcomes, and a continuous gamma response model; it does not fit imaging data. The two implementations agree across 6,804 metric rows (maximum absolute differences 2.090758122186287e-12 for relative change and 4.629630012686903e-13 for correlation) and 201,600 primary predictor values (maximum difference 5.9761223747401004e-15). In 84 runs with flagged delays, changing cue timing produced a median relative predictor change of 5.385104% (25th–75th percentiles 3.884913%–7.800183%), while removing flagged cues produced 38.042338% (28.225970%–52.497842%). Cue removal produced the larger change in every flagged run across all nine settings. These are calculations for fixed schedules and specified models, not a general proof. The task tables do not establish whether a tone played, why a participant responded, or whether a brain signal changed. Sample percentiles describe variation across runs, not population confidence intervals; a second researcher did not check the work.

stop-signal task · cue timing · predictor sensitivity · numerical reproduction · OpenNeuro · control runs · gamma response model · fixed-sample analysis