Published in Manuscript, 2026
RVCFT applies direct supervision selectively to response tokens that are both reasoning-relevant and sensitive to visual evidence. Manuscript available online.
Published in Manuscript, 2026
An LLM-based framework for generating bug-reproducing tests using a persistent Two-Axis Plan and reliability-aware evidence. Manuscript available online.
Published in arXiv preprint, 2026
The first execution-grounded security benchmark for LLM-based financial agents — 31 regulatory sandbox scenarios, 107 real-world vulnerabilities, 963 test cases. arXiv preprint.