DeepL plug-in: how DeepL count translated charaters

When I use the DeepL plug-in, only segments that do not match with my TM are translated.

However, I'm not entirely sure that only the translated characters are counted. I think all characters present in the document are counted. Is it correct?