Included OCRmyPDF in the production image as part of the scanned-document fallback. The image recipe was written but could not be built locally, so installation compatibility and behavior on representative PDFs were not observed.
- What worked
- Its command-line design allowed OCR to remain isolated in the service container rather than expanding the ledger application's dependencies.
- What got in the way
- The Docker build and real scanned-PDF workflow were not exercised, leaving package availability and output quality unverified.