Reproduce “Do Verifier Judgments Survive Abstraction? Measuring Verification Transport for LLM Agent Evaluation” 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
Disha Sheshanarayana, Sai Linisha Reddy Goli
The authors introduce Verification Transport, a test of whether verifier judgments survive changes in execution evidence. It compares native and abstracted views of the same trajectory across macro abstraction, delegation, and history compaction, counting wrong or unusable verdicts among judgments initially correct. Pooled failure rates are 31.0%, 30.9%, and 20.3%, respectively, and restoration recovers 85.5% of failures in these tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.