FAQ
Direct answers for technical buyers.
The product is intentionally narrow. This page covers scope, delivery, fit, and what you should expect from the benchmark.
Is this an OCR benchmark?
It is an invoice extraction benchmark kit for field and line-item extraction, not a generic OCR benchmark.
Is this real invoice data?
No. The invoices are synthetic.
Can I run it locally?
Yes. The evaluator and supporting files ship inside the pack.
What metrics are included?
Per-field accuracy (normalized exact match), line-item accuracy (positional), invoice-level success rate, coverage, and slice summaries by noise, layout, and difficulty. Normalization handles whitespace, date format variants, and numeric rounding to two decimal places.
What is the difference between Starter and Full?
Starter is the main paid benchmark. Full keeps the same workflow and expands coverage.
Why is Starter priced at this level?
Because it saves benchmark setup time and gives you a reusable local evaluation workflow instead of a raw file dump.
Is this suitable for production validation?
It is useful for structured evaluation, comparison, and regression checks. It is not a substitute for validation on your own private documents.
Can I compare two models with it?
Yes. That is one of the main use cases.
What format does the evaluator expect?
The evaluator accepts a JSON file with one entry per document. Each entry contains document_id, header_fields, and line_items. Optional fields for latency and cost are supported. A predictions template and sample predictions file ship inside the pack.
What do I receive after purchase?
A downloadable benchmark kit with PDFs, labels, evaluator, reports, and supporting docs.
Is there a short walkthrough before I request the sample?
Yes. There is a compact public walkthrough with one representative invoice, one sample report view, and one benchmark card view.
How do you handle checkout?
The Free Sample is a direct download. Starter and Full are purchased by email directly from the seller. No checkout platform or account is required.
Who sells this?
Filip Barcík, Cheb, Czech Republic (IČO 24517763). Full operator details are on the imprint page.
Do you offer custom benchmarks or datasets?
Yes. We build custom benchmarks and datasets for any document type, language, or domain. Contact info@ocurka.com with your requirements.
Next step
Start with the free sample.
It is the fastest way to see whether the benchmark format fits your workflow.