The recall contrast remained uncertain after pupil screening in nine-digit trials

We tested whether a fixed pupil quality screen selected memory trials with different recall scores. We used trials with nine digits. We used OpenNeuro ds003838 version 1.0.6. The screen used the right eye, confidence of at least 0.8, and bin coverage of at least 70%. It accepted 1,665 of 2,945 scored trials from 82 matched people. Fifty-four people had enough accepted and rejected trials for a block model. Their mean contrast of accepted recall minus rejected recall was -0.169 correct digits. The 95% interval was [-0.392, +0.051]. We resampled complete people with the observed trials fixed. This assumes independent, exchangeable people in the eligible group. All seven planned checks are included. Only the confidence 0.6 check had an interval below zero. The intervals have no multiple comparison adjustment. An additional check included seven response markers with source scores of zero. It left every eligible person contrast unchanged. The primary interval does not establish equal recall or random data loss. The package includes a reusable audit tool, source files, pupil bins, and trial tables. It also includes all bootstrap draws and retrieval records. Independent checks and offline repeats passed.

pupillometry · working memory · digit span · recall · data quality · trial selection · OpenNeuro · reproducibility