Two gap terms gave an uncertain recall prediction gain under a fixed model in 24 ds005411 test participants

We tested whether two gap terms add recall prediction value for words shown three times. We used 11,790 items from 48 participants in OpenNeuro ds005411 version 1.0.1. A fixed split used 24 participants for model fitting and 24 for testing. The base model used first and last presentation positions and task covariates. The expanded model added signed gap difference and absolute gap imbalance. The mean test log loss gain was 0.000600 natural-log units per word. The 95% interval was -0.001645 to 0.002000 and included zero. The interval used 2000 bootstrap samples of whole participants, with both models refitted. Each participant had equal total weight. The reverse split gave a gain of -0.000182. The test does not establish a reliable gain from these terms under the fixed model. It does not prove that spacing has no memory effect. No brain signal entered a model. The package preserves checked source files, an item dataset, a local plan, code, provenance, predictions, bootstrap samples, and figures. A separate source reconstruction, a second optimizer check, and a complete offline replay passed.

free recall · memory · spacing · repeated words · prediction · logistic regression · bootstrap · OpenNeuro ds005411