Choose the smallest useful specimen
| Task | Safe asset | Measurement |
|---|---|---|
| OCR reading order | An abstract page with numbered labeled regions | Field sequence and region extraction |
| Required-field validation | A fictional JSON record | Correct error placement |
| Date parsing | A dataset of explicit calendar cases | Accepted and rejected inputs |
| Responsive form behavior | A form populated with Sample Person labels | Wrapping, focus and error visibility |
| Demonstration | A plainly fictional data panel | Whether the audience understands the workflow |
An original layout specification
- Use the title “FICTIONAL OCR TEST SHEET — NOT VALID.”
- Place a second permanent fictional label in the body region so a crop retains context.
- Use an abstract issuer such as Fictional Test Authority.
- Use test identifiers containing NOT-VALID, rather than realistic government-number formats.
- Keep issuer seals, flags, official logos, portraits and machine-readable zones out of the specimen.
- Describe each region by its testing purpose, not by a real country's document layout.
Build a measurement plan
Choose the expected extraction result before making the layout. For example, a three-region sheet could contain a Sample Person label, a NOT-VALID reference and a fixed test date. Record the expected text and reading order. Vary one capture property at a time when evaluating an OCR pipeline.
A safe specimen does not establish how an OCR system will perform on every government-document edition. Report that limitation with any result. Label a fixture demonstration as a fixture demonstration; do not market it as an identity-verification benchmark.