Team GT NeuroSignals Lab submitted an ensemble to rank cognitive-impairment risk one to six years after overnight polysomnography within the organizer's time limit. Four feature groups combined demographics with signal summaries, interrupted-sleep measures, stage transitions, and clinical sleep measures. For each group, models trained on all but one data-collection site predicted records from the excluded site. If time allowed, logistic regression combined their probabilities with variables derived from the database record-creation time; otherwise, the program returned the last completed model. Organizer-side training used a deterministic 1,200-record cap. In validation, entry 2029 returned age-conditioned, age-weighted, and overall areas under the receiver operating characteristic curve of 0.748, 0.813, and 0.790, respectively, and a precision-recall area of 0.366. The prevalence-based reward was 0.022, accuracy 0.382, and F-measure 0.167. The organizers did not return the fitted model or training log, so the scored model is unknown. Final assessment requires a single official test-set evaluation of the selected, unchanged procedure.