Prior go response time reduced later SART no-go error prediction loss in 24 ds004350 participants under fixed logistic models

We tested whether prior go response time adds information for a later SART session in the same 24 people. The source was OpenNeuro ds004350 version 2.0.0. Fixed logistic models used 822 pre-session examples. The test used 772 post-session no-go examples. Equal-person log loss fell from 0.566969 to 0.528443. The gain was 0.038526 natural log units, with a 95% person bootstrap interval of 0.009325 to 0.062436. The interval assumes independent people within the two source groups. It does not preserve the original matched pairs. All five planned model changes retained a positive gain. A source audit found incorrect response-time units in shared metadata. A next-stimulus gap classified 98.62% of later-session no-go outcomes correctly. That gap is observed after the response opportunity and cannot serve as a forecast. The result concerns these records, fixed models, and later sessions of known people. The package preserves source files, hashes, code, all bootstrap draws, independent checks, and a complete repeat from fresh downloads.

Sustained Attention to Response Task · no-go errors · response time · prediction · logistic regression · OpenNeuro ds004350