Adding the first interval reduced observed linear prediction error for the third interval by 2.0% in later runs for 23 ds006720 adults after calibration. The 95% interval included zero gain.

We tested whether the first reproduced interval improves linear prediction of the third interval beyond the middle interval. The source was OpenNeuro ds006720 version 1.0.0. Runs 1 and 2 supplied personal calibration. Runs 3 and 4 supplied shared fitting records. Runs 5 through 8 supplied test records. Each shared fit excluded the person being scored. All 23 adults passed the completeness rules. The test used 1,646 complete responses. Adding the first interval reduced observed mean squared natural-log error from 0.184040 to 0.180363. The observed reduction was 2.00%. A whole-person bootstrap with 9,999 refitted draws gave a 95% interval from -0.90% to 5.00%. Thus, this test did not establish a reliable gain. All five specified secondary checks also had intervals that included zero gain. The package contains source files, pinned retrieval records, a local analysis plan, code, predictions, bootstrap draws, figures, and reproduction instructions. An independent numeric-event parser recovered every selected trial. A fresh rebuild matched all 59 analysis result hashes. This is a prediction test after calibration in one public cohort. It does not identify a neural mechanism.

duration memory · behavioral prediction · working memory · linear regression · bootstrap · OpenNeuro