The 95% color shift gain interval included zero in 59 ds008037 v1.0.1 test adults under fixed fits with two kernel widths

We tested a smooth color shift in working memory report models. We used 33,299 reports from 117 adults in OpenNeuro ds008037 v1.0.1. Both models allowed reports near other trial colors and used two kernel widths. We fit each load on 58 people and fixed the fits before scoring 59 people at loads 2, 4, and 6. The mean added gain was 0.000594 nats per report. Its 95% person bootstrap interval was [-0.000333, 0.001524] under the fixed fits. Fitted shifts pointed away from the other colors at all three loads. The fixed reverse split gave 0.001056, with an interval of [0.000302, 0.001838], in 58 test people. The two splits use the same cohort. Other fixed checks address kernel width, absolute color bias, shift bounds, stored degree bins, and prior row inspection. The main evidence permits a small loss and a small gain. It does not show equal models or a neural cause. The public benchmark and its group results were read before the local analysis plan. We preserve the plan, all source tables, fixed source hashes, code, fit attempts, row scores, bootstrap draws, figures, and independent checks. A full replay from fresh source files matched every result table and bootstrap array.

working memory · color report · conditional density · prediction gain · circular statistics · OpenNeuro