Quick answer
Formulas have to be handled as code, not as text. The moment a translator treats SUM as an English word, the spreadsheet stops calculating.
What breaks when a spreadsheet is translated as plain text
- Function names get translated, so
SUMbecomes a word the spreadsheet does not recognise - Argument order is rearranged to suit the target language's grammar
- Cell references are altered or lost, breaking every formula that depends on them
- Named ranges are translated in the formula but not in their definition, so the two no longer match
- Number and date formats get localised inconsistently, changing what the values mean
- Sheet names are translated while the cross-sheet references pointing at them are not
How formulas are kept intact
- Formulas are detected and frozen before translation starts, so the model never sees them as translatable text
- Formula content is separated from cell text and handled on a different path
- References, named ranges, and sheet names are kept in sync as a set rather than cell by cell
- The original formula structure is restored after the translated text is written back
What to check after translating a spreadsheet
- Open the file and confirm totals still match the source
- Look for
#REF!and#NAME?errors, which are the usual signature of a broken translation - Check sheet names against the formulas that reference them
- Verify date and currency cells, where localisation causes silent rather than visible errors
Bottom line
Only the text should be translated; the calculation layer should come back byte for byte. Use the Excel translator, then verify one known total before trusting the file.