With SentenceWindowNodeParser how does

Window size is the sentences captured around the original sentence

So no token counting is done, chunk size is not used. Just counting the number of sentences on either side

oooh that makes sense