evidence-guided

EviReform 提出了一种证据引导的查询重写方法，用于多跳图检索，将检索请求的修订与证据聚合分离，并在 2WikiMultiHopQA、HotpotQA 和 MuSiQue 上超过了最强基线，分别提升了最多 5.59 Recall@5 和 4.50 F1。

This paper introduces EDGE-OPD, a modification of on-policy self-distillation for LLMs that uses guided rollouts and evidence masks to internalize privileged context without degrading general capabilities, showing success in rare-token identity settings.