Reproduce “When Self-Verification Hurts: Calibrated Answer Repair for Financial Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Celine Guo, Weida Liang, Yujian Fu, Anqi Dang, Yanzheng Jin, Kenji Kawaguchi, Wenya Wang
CAR selectively repairs financial answers by estimating missed requirements, checking relevant evidence and relationships, and bounding edits to targeted content. Across three English-language financial benchmarks and three open-weight models, it recovers 14.6% of initially missed requirements while damaging 2.1% of initially satisfied requirements. These system-level rates include unchanged drafts; Fin-RATE uses gold context, so end-to-end retrieval is not tested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.