AUDITFLOW: Executable Symbolic Environments for Structured Financial Reporting Verification

2026-06-01

AI Summary

This work addresses the challenge of verifying structured financial audits, which rely heavily on symbolic evidence such as taxonomic standards and numerical relationships. The authors propose AuditFlow, the first framework integrating an executable symbolic environment with a multi-perspective multi-agent mechanism. AuditFlow constructs a symbolic reasoning space by combining a static US-GAAP taxonomy graph with a dynamic XBRL filing graph, and employs typed tools to support fact retrieval, rule traversal, and deterministic validation. It further introduces dual-perspective junior agents and a senior agent arbitration mechanism to aggregate evidence and compute credibility scores. Evaluated on the FinMR dataset, AuditFlow achieves a joint audit accuracy of 82.09%, outperforming the strongest baseline by 14.93 percentage points. Ablation studies confirm the critical role of the symbolic environment, as its removal causes performance to drop sharply to 17.91%.

Abstract

Structured financial audit verification is difficult for language-model agents because correctness depends on structured evidence rather than text alone. A model must link reported facts to taxonomy concepts, traverse calculation or dimensional relations, and recompute expected values before applying an audit rule. We propose AuditFlow, a graph-grounded multi-agent framework that separates adaptive search from deterministic verification. AuditFlow builds a symbolic environment from a static US-GAAP taxonomy graph and a dynamic XBRL filing graph, and exposes it through typed tools for fact retrieval, taxonomy traversal, numerical checking, and rule evaluation. Two junior auditors inspect each case from regulatory and evidentiary views, while a senior auditor resolves disagreements and can request further investigation. The final reports are fused through evidential aggregation to produce an audit verdict, expected value, evidence trail, and trustworthiness score. On a FinAuditing-derived FinMR sample, AuditFlow reaches 82.09% joint audit accuracy under GPT-5.5, outperforming the strongest baseline by 14.93 points. Removing deterministic checks drops accuracy to 17.91%, showing that the symbolic environment performs the verification step that the model cannot reliably replace.

Problem

• structured financial audit
• symbolic verification
• taxonomy reasoning
• XBRL
• audit accuracy

Innovation

• symbolic environment
• multi-agent framework
• structured financial auditing
• XBRL
• deterministic verification