Gritworks/synthetic-lab-report-ocr-sample

GritWorks Diagnostics Lab Documents: synthetic document sample

A free preview of the Diagnostics Lab Documents sample pack from GritWorks: a fictional diagnostics case from patient registration and lab requisition through physician authorization, specimen query and final lab report.

Each document comes as a clean PDF and a degraded capture (for example blur, photocopy, coffee stain, skew, crop or low light), with expected field values (JSON) and field bounding boxes. Use it to check how OCR, document parsers and vision-language models hold up when capture quality drops.

Want the whole pack? This preview holds 2 of the pack's 5 document pairs. The complete 10-PDF sample pack is free on GritWorks: get the full Diagnostics Lab Documents pack. Qualified teams can also get the full PDF dataset free after a short call.

What's inside

• patient registration: clean + coffee stain
• lab report: clean + low light

Folder	Contents
pdf/	Input PDFs (clean and degraded). Degraded PDFs are rasterized and have no hidden text layer.
ground_truth/	*.json expected values; *.boxes.json field boxes in PDF points (top-left origin); *_bbox.json normalized vertices and 96-DPI pixel boxes.
images/	Page-1 PNG renders (150 DPI) with metadata.jsonl (document type, condition, expected values) for quick viewing.

Boxes cover selected populated fields, not every label or static word. Expected values do not change when a field is obscured or clipped. No OCR confidence is asserted. Keep the JSON out of model inputs.

Good uses

• Field-extraction evals: compare your extractor's output with the expected JSON on clean vs. degraded pairs.
• Regression tests after a model, prompt or pipeline change.
• Sandbox integration for PDF ingestion without touching real customer documents.

Limitations

• A small preview, not a training corpus and not a validated benchmark. The reference data is not an automated scoring system.
• All people, businesses, accounts and records are fictional.
• Automated redaction alone is not a declaration of HIPAA de-identification. These records are fictional; they do not show how to handle real patient data.
• Third-party names or logos, where they appear, belong to their owners and are not covered by the license.

License

CC BY 4.0. Attribute "GritWorks (gritworks.ai)".

About GritWorks

GritWorks builds secure data access for AI development: GritRender generates synthetic documents, images and audio with known expected values; GritRedact detects and redacts PII/PHI with human review and an audit trail; GritTransform extracts fields from PDFs and images into JSON. Need your own layouts, more volume, other languages or formats (Excel, JPEG, PNG, TIFF)? Talk to GritWorks.

Citation

@misc{gritworks_diagnostics_sample_2026,
  title  = {GritWorks Diagnostics Lab Documents: synthetic document sample},
  author = {{GritWorks}},
  year   = {2026},
  url    = {https://gritworks.ai/resources/free-datasets/diagnostics},
  note   = {CC BY 4.0}
}