Update logbook: Repro - Provable Benefits of RLVR over SFT for Reasoning Models

initial commit