So, as I read it, the Libris Mortis explanation basically admits, "no, D&D morality doesn't actually make any sense; please homebrew whatever answer works for your table for underlying reasons that could produce this illogical mess, where {X,y, z} are evil, but {p,q,r} are not".

So I return to my "darts at a dartboard" answer.

So. Can we get a list of the top 5-10 "this is evil?" And the top 5-10 "but this isn't?!"? (And, to remove any real-world stigma attached to those words, maybe we'll replace "evil" with "chicken"). I'll propose "animating the dead" as chicken, and creating golems as not chicken to start the lists.

Then, once we have the lists, we'll see if anyone can propose an algorithm with which an AI could parse actions into chicken and not chicken.

Lastly, if we all agree on the validity of the algorithm, we'll get a good laugh at what that would say about other things.

Sound like a plan?