Reproduce “MoCA-Agent: A Market-of-Claims Code Agent for Financial and Numerical Reasoning” 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
Abdelrahman Abdallah, AbdelRahim A. Elmadany, Sameh Al Natour, Hasan Cavusoglu, Adam Jatowt, Muhammad Abdul-Mageed
MoCA-Agent introduces a multi-agent system that utilizes a "market-of-claims" protocol to verify atomic facts, formulas, and units underlying generated Python programs, thereby mitigating silent miscomputation in financial and numerical reasoning tasks. This framework achieved state-of-the-art results across diverse benchmarks, including an 11.06 percentage point improvement on FinChart-Bench (Open QA) and a 14.39 percentage point improvement on MultiHiertt.
Key insights
- Granular, claim-level verification by specialized agents enables a transparent audit trail of factual, formulaic, and unit-level assumptions, improving robustness against silent miscomputation.
- A structured 'market' protocol allows diverse agents to collaboratively validate sub-components of reasoning, offering a more effective conflict resolution and evidence aggregation mechanism than whole-answer debates.
- Integrating domain-specific, code-aware structural verification with market-grounded repair provides targeted feedback, directly addressing common error patterns in high-stakes numerical reasoning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.