FinAutoRubric: Expert-Guided Automatic Rubric Generation for Evaluating Financial Research Agents

2026-09-28

AI Summary

This study addresses the high cost, poor scalability, and limited institutional adaptability of manually crafted rubrics for evaluating financial research agents by proposing the first expert-guided framework for automated rubric generation. The method employs a collaborative multi-agent architecture comprising writer and reviewer agents, integrated with code execution verification, long-horizon iterative reasoning, and human-in-the-loop mechanisms to enable automatic generation, auditing, and refinement of rubrics while supporting task bank reusability. Experimental results demonstrate that the generated rubrics achieve strong alignment with human scoring across three benchmarks and are preferred in blind evaluations conducted by internal analysts. Furthermore, this work introduces FinAutoRubric, a dataset comprising 100 queries designed to facilitate future research in automated evaluation for financial AI systems.

Abstract

Evaluating finance research agents requires rubrics that reflect expert standards and fix the values correct as of an information cutoff. Expert-reviewed finance benchmarks rely on fixed, per-item rubrics, which are costly to extend and cannot encode each institution's own standard. In FinAutoRubric, experts specify reusable evaluation guidance, while agents and code carry out query-specific rubric generation, review, and validation. This expert guidance governs every agent, as prompts and as rules that code enforces, and a Task Bank of reusable criteria carries it across tasks. In long-horizon loops that follow the expert guidance, a writer agent researches every expected value and a reviewer agent verifies it, and failures escalate to a human. On three expert-authored finance benchmarks, its rubrics track expert scoring as closely as the strongest evaluated generator while stating the expert rubric's expected value for more criteria, their scores agree with human grading, and in-house analysts prefer them in a blind review. The released 100-query FinAutoRubric Benchmark, built from in-house analysts'key questions across 78 tasks and eight asset classes, shows that rubrics from an earlier model generation still leave headroom for a later one.

Problem

• financial research agents
• rubric generation
• evaluation benchmarks
• expert standards

Innovation

• Automatic Rubric Generation
• Expert-Guided Evaluation
• Multi-Agent Collaboration
• Financial Research Agents
• Long-Horizon Loops