papd()

kpe.papd(X, A, Y, propensity, *, n_shuffles=100, n_folds=6, contextual_policy='policy_tree', best_arm='ips', reward_model='random_forest', policy_features=None, n_jobs=1, random_state=None, verbose=False)

PAPD (Population Average Prescriptive Difference), centered, 3-fold.

Implements Eq. S18 of Li & Brunskill (2026): a pure IPW statistic with (Y - mean(Y_test)) centering, no reward model. reward_model is accepted for API symmetry with :func:kpe and is ignored. n_folds is ignored — PAPD always uses 3 folds, the paper's canonical choice.