# Reproduction: Provable Benefits of RLVR over SFT for Reasoning Models: Learning to Backtrack Efficiently

## Pages

| Page |
| --- |
| [Executive summary](#/executive-summary) |
| [Claim 1 — Golden-path SFT](#/claim-1-golden-path-sft) |
| [Claim 2 — RLVR backtracking](#/claim-2-rlvr-backtracking) |
| [Claim 3 — Inference separation](#/claim-3-inference-separation) |
| [Claim 4 — Search agent](#/claim-4-search-agent) |
| [Claim 5 — Trace distillation](#/claim-5-trace-distillation) |
| [Methods](#/methods) |
| [Negative controls](#/negative-controls) |
| [Conclusion](#/conclusion) |