"Reproducibility Dataset for \u201cRegime-Aware Selective Commitment against Adaptive Attackers in Smart Environments"

"This dataset supports the manuscript \u201cRegime-Aware Selective Commitment against Adaptive Attackers in Smart Environments.\u201d It contains the reproducibility-oriented data products used to evaluate a regime-aware selective commitment framework for adaptive defense in smart environments under attacker learning. The study compares four defender policy families\u2014commitment, Nash, reactive, and static\u2014across a static foundation layer and three attacker-learning conditions: stationary within-regime learning, cross-regime transfer, and drifting-regime adaptation. The deposited package includes the primary experiment configuration, static and learning regime catalogues, transfer and drift manifests, merged main-results summaries, block-specific summary CSV files, and a curated subset of representative run-level traces. The experimental design uses a 30-regime static foundation grid, a 15-regime learning catalogue, three locked transfer pairs, and three locked drift schedules, with main experiments run over 10 seeds and appendix sensitivity sweeps over 5 seeds. Standard horizons are 200 steps, while drifting-regime runs use 600-step episodes divided into three 200-step segments. The dataset is intended for result verification, comparative analysis, and manuscript-to-data traceability, rather than as a raw sensor dataset. It is most useful for researchers studying adaptive cyber defense, smart environments, online attacker learning, transfer learning, nonstationary adversaries, and burden-aware security evaluation."