rlvrbook

A reference book on reinforcement learning from verifiable rewards (RLVR), covering how models can be trained using checkable reward signals from math, code, proofs, tools, and agent environments. It targets a broad audience, from newcomers to experienced researchers, with increasing difficulty across chapters. The book is positioned as a comprehensive, up-to-date resource that complements existing literature, with a focus on practical guidance and frontier research.

Key features

• Chapter TL;DRs
• Searchable web version
• PDF download
• Citations for further research
• Practical verifier design checklist
• Core terminology appendix
• Minimal RL background appendix