RepoReasoner: best frontier model manages 69.1% Pass@1 on cross-file execution reasoning even with oracle context.

arXiv 2607.25996 moves code reasoning from function level to repository level, with ground-truth call chains from dynamic tracing of real pytest runs and LLM-based I/O rewriting to suppress memorization. Call Chain Prediction shows high precision and low recall, and longer contexts don't consistently help because of added noise. Performance drops on rewritten data confirm partial memorization reliance.