skema output reference often fails to map back to a correct row of xdd parquet output

[orm011]

The page+block number combination in the extraction json often do not match the parquet contents:
the textual context of the extracted text does not match the text content of the corresponding parquet row (off by 1 sometimes).

I first checked this report isn't just off-by-1 error on my part: some extractions do match their row.