ESPO: Entropy Importance Sampling Policy Optimization — part 3/3

Cleaned-text code points [30964, 31408). Bibliography omitted here; source.md includes it.

All experiments were conducted in accordance with the applicable institutional and national guidelines on research involving (open-source/public) data. The dataset(s) used in this study are publicly available and were originally released under permissive licenses that allow academic reuse. Our models follow standard practices for large-scale language model training; they may still produce biased, offensive, or factually incorrect outputs.