Bank statement parsers are hard to test. Real data is private, every bank formats differently, and everyone claims "99% accuracy" with no way to verify. The Bankstatemently Open Benchmark is a free dataset of synthetic bank and credit card statement PDFs designed around real parsing challenges. 5 countries, 14 edge cases, fictional banks with realistic formatting. Download, parse, submit to the free evaluation API, and get field-level accuracy scores. MIT-licensed, open-source on GitHub.

AppSignal — Real-time monitoring that helps you ship with confidence

Real-time monitoring that helps you ship with confidence

Hey! I'm Michael, I've been building Bankstatemently (a bank statement converter) for the past 6 months. Here's the thing: I'm doing something a bit crazy. I'm giving away all the edge cases I've found while building my product. The stuff other companies would call their IP. While processing bank and credit card statements, I kept discovering quirks that break parsers. I've affectionately started cataloging them, and I'm at 36 now. Things like credit card charges showing up as positive numbers, Thai Buddhist calendar dates (year 2568 instead of 2025), or transaction dates hiding inside the description text. Why would I give this away? Because the challenges aren't a secret. Solving all of them at once, reliably, at scale is actually the hard part. Real bank data is private, so I've turned these challenges into synthetic statements that deliberately include them. 5 are live now, 10 more coming. Each uses a fictional bank with realistic formatting. No real customer data. I also had some fun designing them. Lastly, there's a free evaluation API. Submit your parsed output, and get accuracy scores back. I dare you to get a 100% 🙂 If you know a parsing challenge that's not in the list, tell me. The next 10 statements will partly depend on what you tell me is missing Thanks!

Love this Michael 💪

thanks for you unwavering support, @angus_rogers 🙏

This is genuinely helpful!

@lak7 ah thanks! Glad to know that