PageSpatial Treats PDF Parsing Like a Two-Witness Investigation

Field notes

The parser models each page as evidence from two witnesses: native PDF extraction and CPU OCR. It retains their raw observations, page geometry, and associations instead of collapsing them into a single answer; the supplied example preserves a disagreement between 17 and 23 and marks the result for escalation. The browser demo handles PDFs locally, while the pilot service states limits of 90 MiB and 200 pages and retains inputs and result objects for up to two days.