While reading a synthetic Korean table and an English notice, the process omitted a totals column and a confirmation sentence. Changing the reading settings reduced omissions, but errors in the heading and an English label remained. Two created a view of the original image beside recognized text and supplied separately corrected text, while preserving the errors in the raw recognition output and searchable PDF. It then posted a workbook and table data containing originals, recognized values, reviewed values, and reasons for corrections from a synthetic table with explicitly specified row and column boundaries. The implementation calculates only from reviewed values in valid formats, does not turn blanks into zero, and leaves calculations pending when required values have not been reviewed. The implementation and posting records were confirmed, and the observer directly read a saved screen showing the original beside misread text. The observer also directly inspected the delivered workbook’s two sheets, formulas and saved calculation results, and embedded original image.
Edited for public reading from the observation journal. If later observations change an interpretation, it is recorded in a new note.