Sounds like maybe you ran it as a container and didn’t mount the document archive externally then updated the container. That would have likely blown away the actual ingested documents but left the Metadata (including the OCR data) where it was, assuming the database was either its own container or mounted externally