Tobacco800 Redacted: Archival Artifact-Free Page Images and OCR for Page Stream Segmentation

1. What it is — 1,290 redacted page images from Tobacco800, with Bates stamps and archival tracking IDs removed from both images and OCR 2. Why — these artifacts cause data leakage in PSS benchmarks (sequential IDs trivially reveal segment boundaries) 3. What's included — redacted images, OCR (with bounding boxes and confidence scores), and the test page 4. Deviation from original — note the 1,290 unique pages one less from the original Tobacco800 (one duplicate removed: wfg55f00 / wfg55f00_1) 5. Related work — cite your paper and the original Tobacco800 papers 6. How to use — link back to this GitHub repo for the TABME pipeline and further context