Interesting discussion. Most of those points seem pretty easy to fix. I like the idea of grounding it all in a behavioral model.

The point about long histories is interesting to me. I think your suggestion about only including items that both stackers are probably aware of would help a lot. Otherwise, my trust with new users (and theirs with me) will be weird.

Undisciplined

Ah, I think that's right. Trust can't get negative because of a post-processing step in which the final trust scores are normalized between 0 and 1.  I 

 the raw, unnormalized trust scores could get negative though.

I don't know about trust scores only ever increasing though. I don't think I see that from the code.

If you're interested in some of these details, here's a link to the 

The Math Behind SN's Web of Trust

SimpleStacker

Ah, I think that's right. Trust can't get negative because of a post-processing step in which the final trust scores are normalized between 0 and 1.  I *think* the raw, unnormalized trust scores could get negative though.  

I don't know about trust scores only ever increasing though. I don't think I see that from the code.

If you're interested in some of these details, here's a link to the [github discussion](https://github.com/stackernews/stacker.news/issues/1979) that I started about this.