Reproduce “Interpreting Coding Agents Through Their Behavioral Trajectories” 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
Ibragim Badertdinov, Alexander Golubev
The authors map tool traces from coding agents into shared behavioral actions and compare workflows across deployed configurations. Among 60 same-scaffold pairs with aggregate solve rates within 1 percentage point, 58 differed on at least three of six metrics, versus 0.78% of same-configuration split-half controls. Aggregate matching does not ensure identical solved tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.