Lipschitz-Based Reinforcement Learning for Response-Time Distributions in Video Game Design

response_times_anonymized.csv — 480 participant-level response-time observations (24 participants × 20 game levels). Each record includes response time, trial outcome, and the six level-configuration variables (object size, color, initial X/Y/Z coordinates, spawn time). Direct identifiers and demographic variables have been removed; participants are labeled P01–P24. analysis_pipeline.R — self-contained base-R script (R ≥ 4.1, no external packages) that validates the data, computes level descriptors (median, Q10, Q90, IQR, mean, Pearson skewness), builds the normalized six-dimensional configuration space, runs the fixed-alpha sequential Lipschitz regulator and the one-pass adaptive Q-table procedure, computes McShane–Whitney predictions for the common target configuration, fits the two-parameter Log-Normal reconstruction, and performs level-wise Log-Normal goodness-of-fit tests. learned_q_tables.csv — archived Q-table output (seed 123) with one row per Feature × State × Alpha combination and the greedy action per state. lognormal_goodness_of_fit.csv — Shapiro–Wilk tests on log-transformed response times per level, with Holm-adjusted p-values. README.txt — data dictionary, reproducibility settings (RNG configuration, seed 123, all hyperparameters), and execution instructions.