[so when we pick heuristics, we're thinking of ... things that are simple to implement, for now. maybe also simple to come up with or reuse, but mostly simple to implement and effective.] [what heuristics have we come up with? - autocorrelation to associate letters with sounds. this could be a direct feedback path. - random noise to update tokenizers - checking parsing using a parsing tool (even if the text fails this, on average it helps) - some kind of likelihoodness, like how correctness might be associated with distance - things we come up with after trying something, an incredibly normal approach - hand-added data ... ] [a basic idea here was using multiple channels of confidence with feedback so as to rapidly increase quality with only a little bit of added structure. if this idea is kind of held nearby, it could possibly help. with the thinking issues, maybe we could do a little, then look at the idea, then do a little more, bopping back and forth, unsure.] [too much for some of us]