Gumi — reading historic newspapers and negative search results
Article date / checked: 2026-10-15 (Asia/Shanghai)

中文：这是一份阅读与来源记录，不是 OCR 性能测试报告。
日本語：読解と出典の記録です。OCRの性能評価ではありません。

CONCRETE EXAMPLE
The Library of Congress published a side-by-side OCR/image screenshot in
its April 9, 2024 post below. Its visible heading identifies:
The New-Hampshire gazette and general advertiser (Portsmouth [N.H.]),
October 13, 1787, Image 1.
The article explains subject -> fubjed using long s and the ct ligature.
I viewed the screenshot and compared the word in its left OCR pane with
the corresponding word in the right newspaper-image pane.
The same post discusses the historic spelling authentick vs authentic.

Source post:
https://blogs.loc.gov/headlinesandheroes/2024/04/new-hampshire-newspapers-chronicling-america/
Original screenshot:
https://blogs.loc.gov/headlinesandheroes/files/2024/04/Picture7.png
Local copy, unchanged:
/images/newspaper-ocr-subject-comparison.png
SHA-256:
c4c5f833ccfb55d83b8bf48df977de4c8331e130cf92fc8127ac32b32673541e
Image credit: Library of Congress. The underlying newspaper is from 1787.
The 2024 screenshot shows the interface and text at that time; it is not
a new screenshot of the current site.

PRIMARY REFERENCES AND WHAT WAS USED
1. OCR definition and acknowledged recognition errors:
https://www.loc.gov/collections/chronicling-america/about-this-collection/technical-information/
The official page's indexed text was available; a direct fetch returned
HTTP 403. This access failure is not a search result and is not evidence
that any newspaper word is absent.

2. Selection, digitization, microfilm, and bibliographic scope:
https://guides.loc.gov/chronicling-america/about-the-collection

3. Current guide to Image w/ Text and downloadable image/OCR formats:
https://guides.loc.gov/chronicling-america/download-and-view
Interface instructions are attributed to this guide, not to an invented
hands-on browser session.

4. OCR reprocessing explanation and before/after examples, April 29, 2025:
https://blogs.loc.gov/headlinesandheroes/2025/04/ocr-reprocessing/

5. Software/data release explanation, August 11, 2026:
https://blogs.loc.gov/thesignal/2026/08/yarasavage-new-ocr-reprocessing/
The HTML was retrieved and read. NDNP-Open-OCR was not installed or run.

6. Website migration, changed index and search-link results:
https://www.loc.gov/ndnp/migration/

SCOPE / 未验证范围 / 未検証の範囲
- No fresh OCR run, search-engine benchmark, sampled accuracy estimate,
  recall estimate, or whole-site hit-count experiment was performed.
- No current query result is inferred from the 2024 screenshot. Other
  occurrences, matching behavior, and updated OCR can change retrieval.
- The family-history/obituary scenarios and suggested conclusion wording
  in the article are hypothetical research examples, not performed studies.
- A failed or timed-out HTTP request is not treated as a successful query
  returning zero hits. A separate API discovery request timed out during
  research; it supplies no evidence about the presence or absence of words.
- The article's research workflow is reasoned advice based on the displayed
  example and official documentation, not a guarantee of finding every item.

SIMPLE RECORD TEMPLATE / 记录模板 / 記録ひな形
Collection:
Newspaper / place:
Issue date / page or image:
Stable item link:
Query date / query words / filters:
Available and missing issues or pages:
OCR as supplied:
Reading checked against image / uncertain characters:
What was manually examined:
What the negative finding covers:
Unexamined range / next source: