I ran some (very limited) AI vs. AI tests. Specifically, I tested the following sets of players: (1) Capria, Garrim Gyr, Tya Kiri, Shekinah, Os-Gabella, Weevil, Pickle and Hyde (2) Auric Ulvin, Mahala, Alexis, Varn Gosam, Cardith Lorda, Ethne the White Note that (1) has 2 good, 2 neutral and 2 evil; and (2) has 3 evil and 3 good. I ran 9 games with configuration (1) and 5 with (2). The game options were fairly standard, but trying to reduce random interference. Notably: "Balanced" mapscript, small, quick, no barbarians/lairs/acheron/orthus, compact enforced, no puppet states. I ran the games until turn 450, whoever lived by then, won. Eliminated players are ranked by time of elimination. Here are the results, with average rank +- standard deviation and number of games won: (1) Tya Kiri: 2.3 (0.68-4.0); 3 + 1/3 games won (3 games won alone and 1 won with 2 others) Garrim Gyr: 2.4 (1.1-3.8); 1 + 1/2 games won Os-Gabella: 3.6 (1.9-5.2); 2 games won Capria: 3.6 (1.5-5.6); 2*(1/2) + 1/3 games won Shekinah: 3.6 (1.6-5.5); 1/2 + 1/3 games won Weevil, Pickle and Hyde: 5.0 (3.9-6.1) (2) Auric Ulvin: 1.6 (0.26-2.9); 1 + 3*(1/2) Mahala: 2.2 (0.9-3.5); 2*(1/2) Alexis: 2.4 (0.45-4.3) 3*(1/2) Ethne the White: 4.0 (2.6-5.4) Cardith Lorda: 4.8 (3.5-6.1) Varn Gosam: 5.2 (4.4-6.0) In (1), Weevil is notably worse than everybody else. This is to be expected, as with "No Barbarians" on, the barbarian trait is a straight -10% research and no bonus. Tya Kiri seemed to dominate a lot of games, not sure what is going on there. (2) strongly supports @swapoer's hypothesis. No good civ ever lasted until the end. Alexis actually was a bit worse than Auric and Mahala, because 2 times she just managed to stay around until the end because the strongest civ didn't finish her off. There are, of course, multiple possible reaseons for this. swapoer's suggestion that the evil civs are better suited for earlier war might be correct. However, especially for Doviello this is kind of their main thing, so I'm unsure if I want to take that from them. Also, in one game Elohim were leading the scoreboard until midgame, only to be utterly destroyed by the Illians later. I think trying to get the good civs to work together is the best way to tackle this for now. I'll see what is possible. Disliking players who start aggressive wars would probably a good thing generally. On the other hand, maybe the evil civs should also work together less. By what I observed, the evil players seemed generally happy to dogpile on single good players. I think the dogpiling itself is fine and in-character, but they should probably be just as eager to do it with other evil players as they are with good players. Maybe I'll just remove the "You are evil" diplomacy bonus, at least among AIs, if that's possible. That should mostly just affect the early game. I'd be happy to here more ideas on this. If someone has too much time on their hands, it would be probably best to run a lot of games and only consider the winner each time. It seems that this is the most consistent metric; very weak civs can be eliminated quite late in some circumstances. For now, I'd advise you to use not too many evil civs at once, or try to find and pick weaker ones. EDIT: I should mention that my tests were done on an unreleased build that already toned down sheaim planar gate spawning, so you might get different results in that regard.