frontier models
One Early Mistake Explained More Than Half the Failed Runs
PivotOPD trained agents both to avoid pivotal errors and to recover from the states those errors created.
Summary
PivotOPD trained agents both to avoid pivotal errors and to recover from the states those errors created.
Across three Qwen3 models, the authors found that more than half of failed multi-turn rollouts contained an early pivotal mistake, and that limited teacher guidance after that turn could often restore success. PivotOPD pairs preventive distillation on the correct action with recovery distillation on the following states. It beat thirteen baselines across three agent benchmarks and raised a Nemotron-3.5 student's SWE-Bench Verified resolve rate by 3.2%. These are task-suite gains, not proof of reliable self-repair in open environments.
Why it matters
PivotOPD trained agents both to avoid pivotal errors and to recover from the states those errors created.
Limits and context
- These are task-suite gains, not proof of reliable self-repair in open environments.
Key claims
PivotOPD trained agents both to avoid pivotal errors and to recover from the states those errors created.
Qualification: These are task-suite gains, not proof of reliable self-repair in open environments.
Evidence: source-2026-10-01-013
Sources
- arXiv preprint 2609.40285arXiv · primary research
Corrections
No corrections have been recorded for this story.