This seems like a great idea which I wish was in higher prevalence - designing a simple algorithm yourself that you can understand easily and which can't be gamed, because it is unique.

This seemed to be the promise of spam filters - you get to train them yourself! The difficulty is that it becomes so complex and vague that you can't undo its biases easily. It also reinforces the past at the expense of closing openings for the future to leak in.

Thanks for the code!