Your example is more a case of poorly doing I18N. It is a mistake to put a variable like number_of_herbs inside of a translation. And it is a mistake to not just use codenames like STR_COLLECT_HERBS.
Thanks, fair point. The changing herb count should be a parameter, and I should have shown that the proposed CSV has a stable key column.
The case I meant is an instruction changing under the same key, e.g. UI_INVENTORY_HINT: “Press Tab to open the inventory” becomes “Hold Tab to open the inventory”, while the translated cells remain unchanged. That would call for review, not an automatic correction.
I also see that gettext already has fuzzy review flags. Does your existing workflow catch these source-text changes, or have you run into a gap when working with CSV exports? I’d rather establish that before building another tool.