A standardized benchmark for evaluating bank and credit card statement PDF parsers. 15 synthetic statements across 3 difficulty tiers, 12 countries, 8 languages, and 41 parsing challenges. 5 of 15 statements are released here (PDFs + metadata); the rest are coming soon — see the full dataset on GitHub for status, or browse it on the website.

Design: ground truth is never in this dataset

This repo ships PDFs and structural metadata… See the full description on the dataset page: https://huggingface.co/datasets/Bankstatemently/bank-statement-parsing-benchmark.