The gamma model calculates too few responses at planned delays less than 0.1 s in 126 ds000030 runs.

A common gamma race model was fitted to STOP trials from four of five participant groups and used to predict the held-out group in turn, using 126 CONTROL scanner runs from ds000030 snapshot 1.0.0. At planned delays under 0.1 s, the held-out data contained 108 responses in 241 trials, while the model expected 50.33847508265513. The observed-minus-expected response-rate difference was 0.23925943949105755; a fixed-prediction participant bootstrap gave a descriptive 95% interval of [0.09408749421432586, 0.37220819359002555]. Four test groups had positive differences, while one group, with six short-delay trials, had a negative difference. At delays of at least 0.1 s, the data contained 1,971 responses in 3,791 trials and the model expected 1999.4826898760518. A sensitivity using observed positive GO times expected 47.656681705633 short-delay responses. An exploratory six-participant screen left 79 responses in 208 short-delay trials and expected 39.89445704258118 after refitting. The interval omits refit, GO-estimation, and delay-history uncertainty; training groups overlap. Actual tone onset is unknown, planned delay depends on earlier responses, and shared STOP parameters omit participant differences. These conditional count results do not establish a biological trigger-failure rate or cause, and they do not refute the target's conditional delay-field result.

stop-signal task · gamma race model · trigger failure · participant split · held-out prediction · planned delay · response counts · ds000030 · model checking · reproducibility