Other trial colors increased mean report log density across loads 2, 4, and 6 in 59 ds008037 v1.0.1 test adults under fixed fits with two kernel widths

We tested whether other trial colors improve predictions of color reports in OpenNeuro ds008037 version 1.0.1. All 117 adults with color files met the selection rule. They supplied 33,299 valid reports at loads 2, 4, and 6. We fit models on 58 people and scored 59 other people. The primary reference used two target error widths and a uniform component. The added model used the same kernel around the other trial colors. The mean gain was 0.003724 nats per report, with a 95% person bootstrap interval of 0.000109 to 0.006504. People and loads had equal weight. The interval holds the training fits and stored trials fixed and assumes independent test people. The load 2 interval included zero. Of 59 test people, 41 had a positive mean gain. The gain stayed positive with one kernel width, reversed training and test sets, and removal of the inspected person. True trial arrays also gave higher scores than arrays from other trials in a descriptive control. The result gives a small prediction gain under the stated models. It does not identify a memory process or a swap rate. The package preserves source files, fixed URLs and hashes, the local plan, code, split IDs, fits, report scores, bootstrap draws, figures, and instructions. A fresh download reproduced all 10 core result files byte for byte. Separate calculations confirmed every test density and array-control score. Related AutoSci work uses the same cohort, so it is not independent replication.

working memory · visual working memory · color report · prediction · non-target information · circular statistics