synthetic cfo Back to synthetic cfo
Real generated output

Download a sample synthetic ERP dataset, before you sign up

SAP ECC, a manufacturer in Germany, one fiscal year of purchase to pay and order to cash at medium fraud, scale 600 (600 base transactions per cycle; its 18 tables hold 12,028 rows), built in evaluation mode: the ground-truth marker columns are withheld from the tables and the answer key ships in full. A 1.1 MB download.

This is one complete synthetic cfo package, generated from accounting rules alone, with fraud planted and labelled. Everything below is read from the actual file, not marketing copy. No account needed to look.

12,028table rows
18ERP tables
38labelled fraud records
18fraud schemes planted
What is inside
00 Executive briefThe whitepaper: the company, the rules that built it, the accounting framework, and the financial statements it produced.
01 Audit evidenceThe forensic report, the SOX control matrix, the tax filing summary, and the ground-truth answer key: every planted fraud, by record, control and vector.
02 DatabaseA reconciled SQLite database in which every key between the tables it ships resolves, measured and printed in the package. Query it the way you would query an ERP extract.
03 Source dataExcel workbooks mirroring the real ERP table architecture, so the data lands where your tooling expects it.
Reproducible, and here is the proof

Every dataset is built from a seed. This one was built from seed 20260801 on engine v1.15.0. Give that engine the same seed and the same configuration and it rebuilds the database and every CSV of this package byte for byte; the PDFs and workbooks here carry a SAMPLE watermark, so theirs differ. That is the guarantee no real ledger can offer: any result you get from it can be checked by regenerating it.

This copy is a watermarked sample: every PDF and workbook is stamped SAMPLE, and it leaves out the reproducibility certificate, because the watermark changes those files' bytes. Your own generated packages carry no watermark and include the certificate. The sample's licence, here and on its listings: Free for evaluation and non-commercial research, including publishing results on it with a citation. Not to be redistributed as files, resold, or used to train or evaluate a competing product. (full terms)

The sample's tables and answer key, built from the same seed and configuration, are on Hugging Face and Kaggle as JSON lines; each listing names the engine version it was built on. A starter notebook on Kaggle scores two detectors on them in the browser. The Python client, command line and MCP server are on GitHub.

Download the package

Tell us where to send it and the download starts immediately.

We'll email you the download link, updates when the sample changes, and occasional product news. Unsubscribe in one click.

By downloading you agree to our Terms and Privacy Policy.

Want one built to your own brief?

This sample is one fixed company. The product is the engine behind it: your industry, your jurisdiction, your ERP, your fraud level, a fresh world whenever you want one. The Developer plan is free: all three platforms, the core four cycles, a scale of up to 1,000 base transactions per cycle, and 5 datasets a month.

Create a free account Try the demo See how detectors score on it Talk to us