Case study / Technical evaluation / Healthcare

U.S. diagnostic services provider

Expanding laboratory requisition OCR test coverage

The organization generated production-faithful laboratory requisitions with known expected values and more than 30 controlled negative variants for repeatable OCR testing.

01 / CHALLENGE

Why the existing data could not move.

Laboratory requisitions vary by template, field combination, completion method and image quality. Fax artifacts, blur, stains, low contrast, repeated photocopying, handwriting and skew can cause consequential extraction errors. Production samples contain PHI and rarely provide enough control over rare combinations, while manually creating and degrading forms was inconsistent and difficult to reproduce.

02 / APPROACH

What changed with GritWorks.

GritRender populates production-faithful requisition templates with synthetic patient, provider, test, diagnosis, specimen, insurance and billing data. It systematically creates real-world degradation variants, preserves relationships across requisitions and specimen labels, and provides known expected values for automated OCR validation.

Results

What the organization can now do.

The outcome is not simply a redacted or synthetic document. It is a repeatable workflow that gives technical teams useful evidence without distributing the original sensitive record.

01

Generated large reusable sets of synthetic requisitions without production PHI.

02

Tested OCR across multiple layouts, test categories and difficult image conditions.

03

Created positive, negative, boundary and exception scenarios on demand.

04

Validated matching and mismatching identifiers across requisitions and specimen labels.

05

Measured accuracy by field, document type and degradation condition.

06

Reproduced failed scenarios consistently for investigation and regression testing.

Operational workflow

How the controlled data moves through the system.

  1. 01

    Generate a production-faithful synthetic requisition.

  2. 02

    Create controlled fax, blur, stain, photocopy and handwriting variants.

  3. 03

    Run each document through the OCR workflow.

  4. 04

    Compare extracted output with known expected values.

  5. 05

    Repeat the same cases across fixes, models and releases.

Next steps

Where the workflow expands next.

  • Expand generation to additional requisition types and specialized workflows.
  • Increase coverage of combined degradations and incomplete data.
  • Define OCR accuracy thresholds for every document type and variant.
  • Integrate generation and validation more deeply into CI/CD pipelines.

Start with the blocked workflow

Make sensitive content usable—on terms your enterprise can defend.

Show us the source content, workflow and downstream users. We’ll identify where redaction, synthetic data or a different control is actually appropriate.

Request a working session