R Shiny dashboard demonstrating validation-first analytics for clinical trial duration forecasting. Random split R² = 0.84 vs time-based R² = 0.04—why validation strategy matters more than model selection.

Reproducible ML pipeline evaluating temporal leakage in Expected Pass Turnovers (xPT) models for football analytics. Compares 4 algorithms (mixed-effects logistic, penalised logistic, random forest, XGBoost) across leakage-inclusive and leakage-corrected feature sets. Supporting code for manuscript under review.