# Invoice extraction test — rubric v1

Dataset: `invoice-01.txt` … `invoice-10.txt` (plain-text invoices), answers in `invoice-extraction-v1-answers.csv`. Written before any tool was run.

## Task given to each tool
From each invoice extract: vendor name, vendor VAT number, invoice number, issue date (ISO), due date (ISO or empty), currency, net, VAT rate, VAT amount, total. Tools get only the `.txt` files.

## Scoring
| Field | Points |
|---|---:|
| invoice_number, total | 2 each |
| vendor, vendor_vat, issue_date, due_date, net, vat_amount | 1 each |
| currency, vat_rate | 0.5 each |
| **Max per invoice** | **11** — dataset max 110 |

Dates count only in ISO format (`YYYY-MM-DD`). Amounts must match to the cent. A confidently wrong total counts −1 (worse than a blank).

## Quirks planted on purpose
Two invoices use `DD.MM.YYYY` dates, one has no due date, one has an early-payment discount line, one repeats the total in words. Report how each tool handled every quirk.

## Also record
Setup time, run time, cost, and whether the tool could take all ten files at once.

## Limitations
Plain text only. Says nothing about scanned PDFs, photos, multi-page invoices or non-Latin scripts.