Simulated Scholar Search (S3): scientific-literature environment and verl-based RL research code for search agents.
-
Updated
Jul 26, 2026 - Python
Simulated Scholar Search (S3): scientific-literature environment and verl-based RL research code for search agents.
Copy inflation in multi-turn search agents: 78-92% of generated tokens are copied from retrieved documents and carry inflated log-probabilities, breaking confidence-based methods. Diagnostic toolkit + Retrieval-Grounded Voting. Findings of EMNLP 2026.
On-policy relabeling for robust search-agent distillation under rollout distribution shift.
Official implementation of Lazy Grounding: Attacking Search Agents with Factual Evidence (EMNLP 2026)
To associate your repository with the search-agents topic, visit your repo's landing page and select "manage topics."