Core idea
Incorrect traces preserve exploratory behavior near the model’s capability boundary, while correct traces tend to be shorter and less reflective. ReNIO uses the student/teacher prefix probability ratio to find pivotal tokens leading to wrong reasoning and weights those trajectories more — keeping OPD’s prefix-training advantage (no full rollouts needed).