program-of-thought

Introduces CreditCardQA, a financial literacy benchmark for numerical reasoning derived from real credit card agreements. Evaluates LLMs under Chain-of-Thought and Program-of-Thought prompting, finding Program-of-Thought yields consistent gains and errors stem from misapplied financial rules rather than arithmetic.

RePoT improves Program-of-Thought by enabling deterministic recovery from invalid actions through checkpoint-based repair, achieving higher success rates across multiple models and benchmarks.

This paper introduces FinAgent-RAG, an agentic framework for financial document question answering that combines iterative retrieval, Program-of-Thought reasoning, and adaptive resource allocation to improve accuracy and reduce costs.