Since ingestion errors won't disqualify the entire file from ingestion, missing chunks in the database are possible. These can easily be identified and should be flagged.
If AI cache is present and valid, the corresponding images or blocks should be identifyable, enabling target resume/retry of ingestion, which could be an additional mode.