The live processor accepts the first 25 pages per request. This route explains how to sample difficult pages and divide longer documents without suggesting that a browser request is an unattended batch system.
The live request limit is 25 pages. The value of this route is a safe batching and review method, not a false promise of unattended long-document processing.
Design the review sample with Sample-Based OCR Quality Assurance for Large Collections before dividing the file. Use Multi-Column OCR and Reading Order Review to keep difficult layouts in sequence when batches are reunited. For a single scanned file or a different output, compare PDF to Text and OCR Document Compare.
Sample, divide, recognize, and merge by page identity
Sample front matter, dense text, tables, inserts, and damaged pages.
Process no more than 25 pages in one browser request.
Retain page identifiers and merge only after weak pages are corrected.
Long-document batching scenario
For a 140-page archive scan, sample front matter, ordinary text, foldouts, tables, and the weakest paper. Divide the document at stable boundaries, then maintain a queue of pages needing rotation, recapture, or manual review. Merge corrected text by original page ID, not by processing order.
Find pages that need a different treatment
| Review point | Likely failure | Validation action |
|---|---|---|
| Sampling | Only easy opening pages are tested | Include tables, inserts, damage, rotation, and small print |
| Batch identity | Output loses its original page reference | Name batches and retain explicit page labels |
| Setting drift | One configuration fails on later sections | Start a new batch when layout or scan quality changes |
When browser OCR is not an appropriate batch system
Use a managed batch OCR system when unattended processing, retries, audit logs, and very large collections are required. Extract native text instead when the PDF already contains a reliable searchable layer.