bol-bench: a bill-of-lading parsing benchmark

25 synthetic single-page bill-of-lading PDFs with exact field-level ground truth, built to benchmark PDF parsers on logistics documents. To our knowledge this is the first public bill-of-lading parsing benchmark — Hugging Face previously had no BoL parser or dataset, and ParseBench (arXiv:2604.08538) contains no logistics documents.

Leaderboard + methodology: okrapdf.com/blog/bill-of-lading-ocr-benchmark — 24 parsers scored (text-layer tools, Tesseract, 15 vision models, 5 hosted APIs).

What's in it

Condition	Docs	Description
digital	13	born-digital PDFs with a text layer
scan	7	office-scanner quality: 165 dpi image-only, mild skew + noise, no text layer
fax	5	115 dpi, heavy skew, blur, ordered dither, aggressive JPEG recompression

Six layout families: classic ocean liner form, modern sea waybill, VICS-style straight BoL, LTL pro-sheet, hazmat straight BoL, NVOCC house bill.

• data.jsonl — one row per document: id, template, condition, pdf (filename), and 21 ground-truth fields (bol_number, scac, carrier_name, shipper_name, consignee_name, vessel, voyage, port_of_loading, port_of_discharge, origin, destination, container_numbers, seal_numbers, trailer_number, pro_number, freight_terms, date_of_issue, total_packages, total_weight_value, total_weight_unit, hazmat_un_numbers). Fields not applicable to a document type are null.
• bol-001.pdf … bol-025.pdf — the documents.
• manifest.json — generation manifest.

Provenance & rights

Fully synthetic and deterministic: HTML templates rendered to PDF, degraded subsets rasterized. All shippers, consignees, carriers, and vessels are fictional; no real carrier's form was copied. Ports and UN numbers are factual reference data. Every container number carries a valid ISO 6346 check digit — useful for testing downstream validation (a misread container ID should fail the check digit and bounce to review).

No real shipment, person, or company appears in this dataset. Apache-2.0.

Scoring convention (as used in the leaderboard)

Exact match after normalization (case, punctuation, corporate suffixes) for IDs and names; set-F1 for container_numbers / seal_numbers / hazmat_un_numbers; 0.5% relative tolerance for weights; multi-format date parsing. Score a field only where ground truth is non-null.

Citation

@misc{bolbench2026,
  title  = {bol-bench: a bill-of-lading parsing benchmark},
  author = {okraPDF},
  year   = {2026},
  url    = {https://okrapdf.com/blog/bill-of-lading-ocr-benchmark}
}