The linear time forecast gives a larger mean absolute GO error in all 89 selected ds000030 participants.

This report tests a two-point linear time forecast as a response-time baseline in a failed STOP task. It uses event tables from OpenNeuro ds000030 version 1.0.0 and a selected sample of 89 participants. In 3,493 GO windows, the linear forecast had larger mean absolute error than the constant forecast in every participant. Mean participant MAE was 295.973 ms for the linear forecast and 108.825 ms for the constant forecast; the mean increase was 187.148 ms (95% participant-bootstrap interval 173.325 to 202.052 ms). In 427 STOP windows, the positive-delay minus zero-delay difference in deviation from the linear forecast was 20.461 ms (95% interval -75.773 to 111.125 ms). The estimate remains uncertain and the interval includes negative and positive values. Sensitivity intervals also include zero. The GO check shows poor prediction by this two-point extrapolation in the tested context; it does not validate a forecast for the unobserved response after a STOP trial or establish that the biological difference between STOP classes is zero. Recorded delay classes do not verify physical STOP signal times. The analysis does not adjust for adaptive delay history, central STOP RT, arrow changes, shared schedules, or clinical group. This is computational reproduction on the same observations, not independent biological verification.

response time · stop-signal task · reaction time · forecast error · baseline · linear extrapolation · OpenNeuro · participant bootstrap · post-error slowing