---
格式版本: 2
标题: "FinRCA-Bench: Benchmarking Evidence Retrieval and Reasoning for Financial AI Systems"
原文链接: "https://arxiv.org/abs/2608.18534"
发布日期: "2026-08-19"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Wed, 19 Aug 2026 04:36:01 UTC (867 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T16:04:58+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T15:56:53+08:00"
入库时间: "2026-08-20T08:04:58.239Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=AI&searchtype=all"
匹配关键词:
  - "AI"
  - "performance"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 10
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "内容为金融领域AI基准测试论文，与超节点、AI Rack、机柜级AI基础设施、供电散热互连等核心主题完全无关，仅因搜索词AI命中，判定为无关。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T16:11:19+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 5
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 0
AI摘要: "论文提出FinRCA-Bench，一个包含2250个应付账款到银行对账案例的合成基准，用于分离金融AI系统中的证据检索与推理。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:19:57.878Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18534"
---

## Computer Science > Artificial Intelligence

## Title:FinRCA-Bench: Benchmarking Evidence Retrieval and Reasoning for Financial AI Systems

Authors:[Pratik Ghawate](https://arxiv.org/search/cs?searchtype=author&query=Ghawate,+P)

[View PDF](https://arxiv.org/pdf/2608.18534)

> Abstract:Large language models are increasingly used to support financial operations, but their apparent reasoning performance can depend on whether they receive the right evidence. In financial reconciliation, the evidence needed for diagnosis is distributed across invoices, purchase orders, approvals, allocations, payments, ledger entries, and bank activity, linked by transactional relationships rather than textual similarity. End-to-end accuracy can therefore conflate evidence access with reasoning quality. We introduce FinRCA-Bench, a deterministic synthetic benchmark of 2,250 accounts-payable-to-bank reconciliation cases spanning 14 operational tables, including 1,500 injected failures across 15 causal categories and 750 legitimate or hard-negative cases. Root-cause labels and record-level evidence contracts are hidden from the model, allowing retrieval to be evaluated independently of answer correctness. We compare Rules/SQL, classical machine learning, dense semantic retrieval, deterministic relational expansion, and Typed Provenance Graph Retrieval (TPGR), a typed traversal restricted to persisted transaction relationships. Rules/SQL reaches 84.97% held-out exact accuracy and classical ML reaches 95.44%. Holding the reasoning model, prompt, and generation settings fixed while changing only retrieval increases macro required-record recall from 0.83% to 77.70% and exact 16-class accuracy from 2.05% to 72.44%. Structural retrieval failures outnumber reasoning failures with sufficient retrieval by 95 to 15; 254 correct predictions occur despite incomplete retrieval, and strict returned-evidence contract accuracy is only 5.72%. On FinRCA-Bench, retrieval architecture strongly shapes observed AI-system performance, and a correct root-cause label is a weak proxy for an auditable diagnosis.

| Comments: |  |
| --- | --- |
| Subjects: | Artificial Intelligence (cs.AI); Information Retrieval (cs.IR) |
| Cite as: | [arXiv:2608.18534](https://arxiv.org/abs/2608.18534) \[cs.AI\] |
|  | (or [arXiv:2608.18534v1](https://arxiv.org/abs/2608.18534v1) \[cs.AI\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18534](https://doi.org/10.48550/arXiv.2608.18534) |

## Submission history

From: Pratik Ghawate \[[view email](https://arxiv.org/show-email/8ff7165a/2608.18534)\]  
**\[v1\]** Wed, 19 Aug 2026 04:36:01 UTC (867 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18534) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))