• 2026-09-27 TheoremSmith: Mass-Producing Hard Proof Problems to Train LLMs