FinBalance: A Multi-Document Accounting Reconciliation Benchmark

2026-06-14

AI Summary

Existing financial NLP benchmarks primarily focus on curated financial texts, limiting their ability to evaluate models’ capacity for accounting reconciliation, journal entry generation, and balance sheet consistency verification using raw, multi-source documents. This work proposes the first end-to-end reconciliation benchmark grounded in real-world accounting workflows, constructing source document packages spanning eight industries, three reporting cycles, and five difficulty levels. A deterministic generator synthesizes realistic business scenarios, accounting policies, tax and foreign exchange treatments, along with 23 types of inconsistency labels, and automatically produces corresponding journal entries and balance sheets. Experiments show that state-of-the-art large language models achieve only a 46% perfect accuracy rate on balance sheet generation across 710 test instances. Ledger replay analysis further reveals a 26–41 percentage point gap between model-reported and ground-truth reconstructed results, with domain experts confirming the benchmark’s validity and challenge.

Abstract

Existing financial-NLP benchmarks mostly evaluate prepared artifacts such as filings, tables, or extracted values. Real accounting begins earlier: source documents must be reconciled into cited journal entries, aggregated into a balance sheet, and checked for contradictions. We introduce FinBalance, a multi-document accounting reconciliation benchmark built from source-document bundles across eight industries, three period types, and five difficulty levels. Human-authored business scenarios, accounting policies, tax/FX treatments, document schemas, distractors, and inconsistency templates are composed by a deterministic generator whose ledger produces journal entries,balance sheets, and 23 inconsistency-code labels. On a 710-record evaluation split, six contemporary LLMs reach at most 46% exact final-balance-sheet accuracy. Four models show a 26-41 pp gap between BS_exact, the model's reported balance sheet, and BS_recon, the balance sheet obtained by replaying its entries through our ledger. Models often recover numerically plausible entries but fail to bind them to supporting documents and aggregate them consistently. Citation-pressure prompting barely changes document-linking errors, while ledger-feedback ablations substantially improve reported balance sheets and expose inconsistency-detection trade-offs. Expert finance reviewers validate the benchmark design and labels.

Problem

• accounting reconciliation
• multi-document reasoning
• financial NLP
• balance sheet consistency
• source document alignment

Innovation

• accounting reconciliation
• multi-document reasoning
• financial NLP benchmark
• ledger-based validation
• inconsistency detection