MoCA-Agent: A Market-of-Claims Code Agent for Financial and Numerical Reasoning

Researchers introduced MoCA-Agent, a novel AI system that improves financial and numerical reasoning by decomposing questions into atomic claims verified through a market-based mechanism rather than free-form debate. The system achieved strong performance across ten benchmarks, including 78.3% on FinQA and 86.9% on ESGenius, demonstrating that claim-level verification enhances accuracy in high-stakes numerical reasoning tasks.

MoCA-Agent addresses a critical vulnerability in AI systems handling financial data: the tendency to produce plausible but factually incorrect answers when reasoning over tables, formulas, and numerical values. Traditional multi-agent debate systems lack mechanisms to ensure answers ground properly in source data. This research applies market mechanisms—typically used in prediction markets—to constrain AI reasoning, where specialist agents place bids on atomic claims, creating confidence-weighted decisions backed by verifiable evidence.

The innovation reflects growing recognition that financial AI requires structural guarantees beyond language fluency. Errors in unit conversions, cell readings, or arithmetic operations can compound silently in financial contexts. By breaking problems into typed atomic claims and synthesizing Python code only from market-cleared evidence, MoCA-Agent introduces auditability and consistency checks—critical for regulated financial applications.

The benchmark results across diverse domains suggest this approach generalizes beyond niche problems. Performance gains on FinQA, FinanceMath, and FinChart-Bench indicate real-world applicability for financial institutions, investment platforms, and compliance teams that rely on AI-driven numerical analysis. A code-aware verifier layer catches structural and logical errors before execution, further reducing failure modes.

This work signals industry movement toward explainable, verifiable AI for financial reasoning. Future developments may integrate these mechanisms into production systems, particularly in regulatory reporting, ESG analysis, and quantitative research where auditability directly impacts institutional risk. The open-source release enables broader adoption and refinement of claim-based verification techniques across domains requiring numerical precision.

• →MoCA-Agent uses market-based claim verification instead of free-form debate, improving accuracy in financial AI reasoning.
• →The system decomposes questions into atomic claims, verifies them through specialist agent trading, and synthesizes executable code.
• →Achieved 78.3% on FinQA and 86.9% on ESGenius, demonstrating consistent performance across ten financial and tabular benchmarks.
• →A code-aware verifier checks programs for execution errors and common financial reasoning mistakes before final output.
• →Open-source release enables broader adoption of claim-level verification for high-stakes numerical reasoning applications.