We need a better way to manage tokens in a document kv class. Currently there is only one perspective because the raw object values are used to store the tokens, so we can't use a bump operation or something like it to store a completely alternative list of tokens.