One of my focuses of this project is to make sure everything is as heavily scrutinized as possible while still being able to automate. I think the prompting and fallback logic I'm using has worked very well so far. There were some kinks at first that had to be worked out, and I'm very actively fixing anything that arises. As far as being unbiased, you can decide on your own by reading. I think it's doing pretty well.