soham27's picture
Release Banking AML Smurfing & FinCEN SAR Audit Bench v1.0
cd7e701 verified
|
Raw History Blame Contribute Delete
1.74 kB
metadata
license: apache-2.0
task_categories:
  - question-answering
  - table-question-answering
  - zero-shot-classification
language:
  - en
tags:
  - enterprise
  - rag
  - evaluation
  - benchmark
  - banking_and_compliance
size_categories:
  - 1K<n<10K

Banking AML Smurfing & FinCEN SAR Audit Bench

High-Consequence Enterprise Evaluation Benchmark

Official Maintainer: Botmartz AI Research (botmartz.com)
Research Publication: botmartz.com/insights
Primary Metric: SAR Multi-Hop Structuring Detection Precision

Problem Overview

In high-consequence enterprise applications within FinTech Compliance, standard out-of-the-box LLMs and naive dense vector RAG fail on multi-hop dependencies, tabular constraints, and strict regulatory terminology.

This benchmark provides rigorous ground truth for evaluating:

  • Retrieval precision across dense, sparse (BM25), and graph architectures.
  • Structural layout and AST preservation.
  • Deterministic extraction accuracy on mission-critical documentation.

Data Splits & Files

  • train.jsonl: 350 calibration records with rich metadata, Markdown ASTs, and ground-truth QA pairs.
  • test.jsonl: 120 blinded evaluation records.
  • records.csv: Full tabular catalog with clean schemas.
  • baseline_model_eval.csv: Empirical benchmarks comparing Botmartz Deterministic Pipeline vs. Naive Vector RAG.

Citation

@dataset{botmartz_banking_aml_structuring_bench,
  author       = {Botmartz AI Research Team},
  title        = {Banking AML Smurfing & FinCEN SAR Audit Bench},
  year         = {2026},
  publisher    = {Botmartz AI Solutions},
  howpublished = {\url{https://www.botmartz.com}}
}