Team GT NeuroSignals Lab developed a stacked ensemble as part of Screening for Cognitive Impairment During Sleep Studies: The George B. Moody PhysioNet Challenge 2026. The approach combines demographics with signal summaries, interrupted-sleep measures, stage changes, and clinical sleep measures. Base classifiers predict records from sites excluded during fitting. Logistic regression then combines their probabilities with recording-time variables. A deterministic 1,200-record sample limits training cost, and intermediate models remain available if training exceeds the time budget. Retrospective leave-one-site-out evaluation of the complete stack yielded age-conditioned areas under the receiver operating characteristic curve (AUROCs) of 0.731, 0.644, and 0.653 for I0002, I0006, and S0001. Our entry received an age-conditioned AUROC of 0.748 on hidden validation data. The organizers did not return the fitted model or training log, so the score cannot be attributed to the complete stack or a particular intermediate model. The hidden test score and final rank were not yet available.