Hey, I'm Zach, the author of this. I had this idea as soon as Jev appeared: how would Jev and other decision models handle the Trolley Problem? So I ran Jev and 13 other models through many different versions of the question.

I used resampling and multiple variations and orderings of each question so a random lucky phrasing wouldn't dominate the result.

Happy to chat about the philosophical side of the article, strategies for prompting decision models, or anything on the methodology.