Reproduce “Trace-Level Analysis of Information Contamination in Multi-Agent Systems” 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
Anna Mazhar, Huzaifa Suri, Sainyam Galhotra
Researchers from Cornell University and University of Illinois Urbana-Champaign developed a trace-level analysis method to systematically study information contamination in multi-agent systems. Their work reveals that errors from extracted data can propagate silently, leading to incorrect outcomes without workflow structural changes, or cause expensive detours that still recover to correct answers.
Key insights
- A significant decoupling exists between structural workflow divergence and final task outcome correctness; errors can propagate silently without triggering control-flow changes, or workflows can incur costly detours yet still recover to correct answers.
- Contamination manifests in modality-specific control-flow patterns, such as early termination for audio transcription failures or strategy rerouting for tabular data issues, indicating localized vulnerabilities.
- The timing of the 'first divergence point' in an execution trace provides critical insight into the failure mechanism, localizing issues to initial extraction failures or later reasoning-stage sensitivities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.