How data moves

bloody-level keeps the import path local and inspectable:

PDF file
  -> SHA-256 identity check
  -> PDFium text extraction
  -> optional local OCR
  -> deterministic parsing and analyte Library mapping
  -> encrypted vault and audit events
  -> report, patient, chart, and CSV views

The original PDF is copied into application data. Extracted text, canonical rows, inline prior rows, parser diagnostics, and audit information are stored with the report so a later review can distinguish source data from interpretation.

What the app does not do

• It does not upload reports to a required cloud service.
• It does not silently replace a failed parser with an invented value.
• It does not treat an unmatched line as a confirmed analyte.
• It does not turn a reference range or trend into a diagnosis.

Optional OCR and model-backed paths remain local and are only used when configured and available. Their status should be checked in Settings and in the report diagnostics.

Trust the source in the right order

When something looks surprising, check the original PDF first, then the report diagnostics, then the canonical row and chart. A chart is a convenient view over stored data; it is not a replacement for the source document.