What it is
Quick start · Metrics · Python API · Run on Kaggle
An invoice parser can read 1,234.56 correctly and misread 1.234,56. This Python CLI makes that regression reproducible: render the same synthetic
invoice data as English, German, and Spanish PNGs, run your extractor, and
compare field accuracy with deterministic CI gates.
How it works
On September 29, 2026, OpenAI introduced GPT-6.1 Sol, highlighting document understanding and professional workflows. That motivates a narrower deployment question: does your extractor preserve accuracy when invoice conventions change? This package supplies the experiment, not evidence of a defect in that model.
Results
| field_accuracy | EN | DE | ES |
|---|---|---|---|
| invoice_number | 100.0% | 100.0% | 100.0% |
| invoice_date | 100.0% | 100.0% | 100.0% |
| net_amount | 100.0% | 0.0% | 0.0% |
| tax_amount | 100.0% | 0.0% | 0.0% |
| total_amount | 100.0% | 0.0% | 0.0% |
Try it
pip install locale-invoice-check
locale-check demo

