After editing the translation, I checked how 337 pages read

I began with about ten comparisons of the source, existing translation and proposed revision. Separating missing meaning, unclear references, specialist concepts, awkward wording and OCR errors gave me a more useful editing brief than simply asking for a better translation.

I chose Korean terms with the original alongside them and explanatory notes where needed. Lettering inside illustrations stayed intact; external captions were translated. I corrected the rule that mistook short captions for embedded labels by checking image boundaries and position.

Word was useful for editing, but did not reproduce the source’s complex columns and illustrations well enough. I used HTML for comparison and PDF for print-layout inspection, with JSON checks and a report tying the output to its review.

OCR had merged left and right columns, leaving text over illustrations and empty explanatory panels. I separated columns, adjusted difficult pages individually and restored 19 grey panels. The final structural check covered 337 pages and 950 text boxes, with zero missing assets, boundary violations, box collisions or rendered overflows.

The retained PDF matches the hash in the final check. That connects the file to its layout results, but I still need to review architectural meaning and explanation. Six bibliography pages remain source images, limiting text search and selection.

I would again begin with language samples and written rules for body text, captions and embedded lettering. Then I would check pages, assets, boxes and actual rendering separately, using shared rules for repeated errors and individual adjustments for unusual pages.

I bring together language samples, editing decisions and layout checks. The 337-page figures describe structure, not translation accuracy.