Update logbook: Repro - On the Power of Approximate Reward Models for Inference-Time Scaling

initial commit