HTR/OCR recognition
| Then | No automated reading existed; transcription was entirely manual. |
|---|---|
| Now | Handwritten text recognition and OCR produce candidate text with character errors. Output must be corrected against the image by a human. |
| What remains uncertain | Error rates on damaged, unfamiliar or idiosyncratic hands, and how generalisable a model is beyond its training data. |
| Reliability | Useful as a first draft; not trustworthy until human-corrected. |
| Source date | 2026-10-17 |
Research anchors
- ANSI/ASB Standard 070, Standard for Examination of Handwritten Items
- NIST/NIJ, Forensic Handwriting Examination and Human Factors
- NIST IR 8282, Handwriting Examination: Meeting the Challenges of Science and the Law
- US National Archives, OCR transcription validation
- Library of Congress, Paleography guide
- State Archives of North Carolina / DoHistory, How to Read 18th Century British-American Writing
These sources inform the lesson. They do not qualify this site or its readers as forensic examiners.
All methods