Currently, all character tokens enter the tree builder through the one method that does various checks. If we are able to keep complex feedback from tree builder into the tokenizer, the tokenizer could transition into a fast-track 'data state' when the tree builder is in a mode that just ends up writing through to and accumulation buffer. This way, the fast-track 'data state' could bypass the conditionals for the more complex tree builder modes. This won't work if we end up doing speculative tokenization as opposed to speculative tree building.