That is the way it has to work. Remember that this is not a spell check in a plain text file. So when scanning for errors, calibre has to parse the html and store locations for misspelled words only between html tags and in some types of attributes, such as title and alt. That means that in order to present accurate counts and allow jump to location on double click, the book has to be rescanned if you edit outside of the spell check tool. So dont do that. Put the spell check window to one side and look at the words in context, but actually make corrections only in the tool.
|