Previous responses reduced later log duration forecast error by 7.57% in 23 ds006720 adults with personal linear fits

We tested whether the previous timing response added forecast value in OpenNeuro ds006720 version 1.0.0. We fitted personal linear models on runs 1 to 4 and scored valid response pairs on runs 5 to 8. All 23 adults met the fixed selection rules. The base model used current and previous condition labels and task position. The full model also used the previous log response duration. Mean squared log error fell from 0.126788 to 0.117194. This was a relative gain of 7.57%. The 95% interval was 3.37% to 11.34% from person resampling with model fits. The interval assumes independent people. The gain was 3.55% without position terms. The check with training on six runs had a 1.39% gain with an interval that included zero. All nine planned comparisons are reported. Independent event and regression checks passed. A fresh download and full repeat matched all 64 generated result files byte for byte. The package preserves events, source hashes, code, models, forecasts, draws, and repeat instructions. The result applies to complete responses after personal calibration within one task visit. It does not identify a neural or causal mechanism.

duration reproduction · sequential dependence · forecasting · working memory · personal calibration · OpenNeuro