Pairwise Draft Ranking Model

Instead of predicting an absolute outcome (like career wins), build a model that predicts which of two prospects will have the better career and aggregate those head-to-head probabilities into a ranking. Augment that approach with human‑curated intermediate labels (role archetype probabilities) so the model evaluates players relative to likely NBA roles rather than raw box‑score outputs. — This design is a replicable pattern for reducing noise in predictive tasks where long‑run outcomes are heavily influenced by luck or context, and it highlights the value of hybrid human–machine pipelines.