Document classification and splitting: synthetic manifest reference
Run: python3 check.py
Run: python3 split_reference.py
No external dependencies or calls. No PDFs or personal data included.
The validator checks one owner for every original 1-based page, declared classes and ascending page sets. It does not prove semantic correctness, source authenticity, file integrity or routing permissions. Boundary metrics require contiguous document instances; coverage validation can represent noncontiguous page sets.
The merged candidate retains all page labels but merges two invoice instances. Correct exact matches are REC-A, CR-A, UNKNOWN-A and BLANK-A: four of six reference instances.
Original AGI reference material for internal evaluation and adaptation. No warranty of production fitness.
