Replacing zero stop delays with planned delays reduced the mean gap between training and MRI estimates in ds000030 version 1.0.0 under the stated integration rule

We tested whether delay policy changes agreement between training and MRI stopping-time estimates. We audited 533 sessions from 267 participants in OpenNeuro ds000030 version 1.0.0. There were 941 STOP rows with zero recorded delay and positive planned delay. Every such row contained a response before the planned delay. We replaced these zeros with planned delays as a sensitivity test. Under the stated integration rule, this lowered the mean estimate by 16.5 ms in training and 32.8 ms in MRI. Among 237 pairs that passed the stated quality screen, the mean MRI minus training gap fell from 25.8 ms to 8.6 ms. The replacement gap had a 95% bootstrap interval of 0.6 to 17.2 ms. Descriptive agreement limits were minus 118.2 ms and plus 135.4 ms. Thus, the smaller mean gap did not imply close agreement for each participant. An independent comparison matched all 1,596 finite policy estimates in a related AutoSci publication. This reproduces its delay finding and adds task software inspection and absolute agreement estimates. The package preserves source bytes, hashes, code, results, tests, and reproduction instructions. A repeat analysis matched all main output files. Planned delay does not prove sound onset, and replacement is not a verified correction.

stop-signal task · stop-signal reaction time · response inhibition · delay policy · measurement agreement · OpenNeuro ds000030