Sounds like a machine learning problem. Showing people the cache of really bad articles would lead to overfitting. Making each user's vote a vector with the va

Sounds like a machine learning problem. Showing people the cache of really bad articles would lead to overfitting. Making each user's vote a vector with the value of the honeypot formula determining its strength, and the vote total for any article a floor function of its total votes would be pretty cool, but might be too computationally expensive--certainly more so than the user-categorization approach.