The archive
Statistics
The source, the machine reading, the reconstructed structure and the finished document, in figures.
Source
| File | The China Affair manuscript (scanned 7-26).pdf |
| Pages | 267 |
| Size | 52.6 MB |
| Text layer | none — pure raster scan |
| Page size | 792 × 612 pt (US Letter, landscape-flagged) |
| Producer | EFI Cyclone |
OCR
| Engine | Tesseract 4.1.1, --psm 3, English |
| Render | pdftoppm -gray -r 300 (2550 × 3301 px/page) |
| Output | TSV with per-word bounding boxes and confidence |
| Lines recognised | 7,740 |
| Median line confidence | 95.2% |
| Pages ≥ 90% mean body confidence | 261 / 266 |
| Pages 80–90% | 5 |
| Pages < 80% | 0 |
PSM 3 was chosen over 4 and 6 after comparing all three on sample pages: it preserved the blank lines between paragraphs and avoided a spurious leading character that PSM 6 introduced.
Reconstructed structure
| Block type | Count |
|---|---|
| Body paragraphs | 522 |
| Chapter headings | 15 |
| Footnotes | 31 |
| Block quotations | 6 |
| Verse / telegram blocks | 5 |
| Total words | 67,914 |
Footnote separator rules detected from pixel data on 30 pages.
Output
| Format | .docx, US Letter, 1"/1.25" margins |
| Type | 12 pt Times New Roman, double-spaced, 0.5" first-line indent |
| Length | 227 pages |
| Footnotes | 31, as real Word footnotes anchored at the author's asterisks |
| Validation | passed structural XSD checks |