Public health receives immunization-clinic packets that mix born-digital PDFs with scanned pages and still needs structured clinic fields. What should the extraction design support?

Public health immunization-clinic packets that mix born-digital PDFs with scanned pages require field extraction designed to work across that mix. Clinic demographics, vaccine codes, lot numbers, and consent marks may appear on digitally generated pages in one packet and on scanned wet-ink pages in the next; the pipeline must not assume a single input modality. Table extraction across mixed digital and scanned immunization-clinic packets captures tabular grids but does not extract structured clinic fields like demographics, vaccine codes, and lot numbers outside tables the stem requires. Digital-PDF extraction across mixed digital and scanned immunization-clinic packets processes only born-digital pages and leaves scanned pages unhandled so structured clinic fields on scans are lost for reporting and follow-up. Scanned-page extraction across mixed digital and scanned immunization-clinic packets processes only scanned pages and leaves born-digital pages unhandled so structured clinic fields on digital pages are lost for reporting and follow-up. Exam caveat: mixed digital-plus-scanned municipal clinic packets need unified field extraction across both page types—not table extraction that captures only tables, not digital-PDF extraction that handles only digital pages, and not scanned-page extraction that handles only scanned pages. Operational check: sample packets with both born-digital and scanned pages; run one extraction design across both; verify critical clinic fields populate regardless of page origin; escalate low-confidence scans for human review before immunization records finalize.