I'm even more clueless about the the field.

Those two types of reasoning, are those well definitely in the field generally, or are they mini-personal evals you run?

It has been frustrating trying to quantify why one pseudo-frontier model seems better at some non-coding tasks than another pseudo-frontier one.