Sullla's AI Survivor Season Seven - Playoff Game 1 Thread

[Saxo Grammaticus]

Playoffs continue Friday, June 30th on Twitch at noon EDT, 5pm BST! Kindly note the one-week break, and stay tuned for Alternate Histories on the 23rd.

You can catch the Wildcard Game on Twitch and YouTube.

If you missed basically all of the season, this Playoff may look a little odd We have one seeded leader along with a number of middling ones including two who were previously ranked 49 and 51... Given how little exposure we have had to how half of the leaders perform, this could turn out to be a very interesting game.

Watch the preview here, read up on the game here, and make your predictions here. All are welcome to discuss the game in this thread and follow along for what will hopefully continue to be a dynamic and entertaining season!

And if one contest just isn't enough, odds are already circulating in the back alleys of CFC--rumor has it @Fippy is the one to talk to, but you didn't hear it from me...

 

[AcaMetis]

Part of me is like "this is where the curse of season seven will be broken and maps return to normality". Another part of me is like "with a crowd like this what even is normality?"

 

[elitetroops]

This map suffers from a severe plains cow shortage. Seems there's only one south of Sury and another east of Rammy. Will either of them grab it with their second city? And could that save Rammy from the crazy featherhead to his west?

 

[Keler]

I really think some AI leaders like Ramesses,Saladin,Suleiman,Qin Shi Huang are very underrated among seeded leader fans . And somehow I once ended up ranking Ramesses second most victorious after HC. Overall he wins far more often than many other leaders, maybe except Pacal too. I don't know. I still think this egyptian leader is the best AI among these six if not then Suryavarman, who appears to be worst enemy of half of the world. Very unlucky on his part to end up in a game like this. And Joao is overall nerfed without tech trading asn AI but has great start and usually a good AI. Actually both his opening round and playoff game starts very identical too.
I will pick
Ramesses to win simply because I want him to earn points in the ranking and reach to spot where he deserves.
Churchill to second place for trolls and fun, and huge peninsula next to him.
Montezuma for first to die, hoping english army cross walk and attack him, assuming he can't plot anyone else.
316 turn 9 wars.

edit: forgot to say spaceship because instead of culture in case ramesses doesn't win

 

[elmurcis]

Monty has 6 FP and true power start - auto win!

 

[ManiaMuse89]

Dreading picking for this one. All of the leaders are bad imo.

Sury is overrated I think. A JC-lite kind of leader but even more inconsistent.

Hammy is a pushover. Even when he gets big he is prone to spontaneously imploding from a war dec unless someone joins in to save him.

I don't really know what Joao does with his traits but often the results are very 'meh' and middle of the field.

Churchill is completely passive and very unlikely to do anything to try to win the game. Fairly good at defending himself I guess but that doesn't help him win.

As is Ramesses and he is worse than Hatty at going for culture. War chariots are good but it looks like his start does not have horses.

Monty is Monty. Likely to die first and get dogpiled for being too aggressive and the wrong religion although always the outside chance of an amusing snowball win.

I doubt there will be too many twists. War declarations will probably result in quick eliminations. Strong possibility of a very boring and peaceful ending if everyone is pleased (only Sury and Monty plot at pleased and Monty is likely to be dead by the end of the game).

Religious blocks may be particularly important I think.

 

[elitetroops]

I might vote for Monty to continue the memes. He does have an excellent start with 100% riverside BFC and not too many forests, which tend to slow AI down. No early copper means no stupidly early war and once he gets construction he also has elephants. Ramesses will not have elephants and there's a good chance he can't even pick up iron. All early Egyptian wonders will soon be Aztec when Ramesses is FTD.

Meanwhile the expansionists Joao and Sury will have built up a ton of border pressure and will be fighting a war that doesn't give any of them any advantage. Eventually Monty joins in on either side and grabs a ton more land.

Church will sit in his corner and be Churchill. Not gonna do much all game, until quite late he decides to roll over Babylon to secure a 2nd place finish, as Monty wins domination t299.

Btw. how do the playoffs work? Top 2 from each game proceeds to finals, rest are out for good?

 

[Slashninja1467]

Just to illuminate because i havent seen this discussed but the two factors related to whether an ai will pursue cultural victory are traits and # of holy cities. Credit to Aethylwind in twitch chat for explaining it right here:

"In particular, the AI progresses to the next "culture victory stage" 1 era later if the X/Y values of its capital city add up to an odd number than if they add up to an even number.
It has to value culture at 10 in order to pursue cultural victory, it starts at -3 because of Aggressive AI, gets 4 points for any of the traits Spiritual, Creative, Philosophical, Financial, and Industrious, 1 point for every holy city, and a random value from 0-6 depending on the location of its capital city. There's also another -3 I can't identify related to the AI's favorite civic - my best guess is Theocracy penalizes them another -3 (It's coded as isNoNonStateReligionSpread, which would suggest Theocracy).
So with only one Culture-valuing traits, you'd always need at least 3 holy cities to pursue cultural victory (-3+4+3+6 = 10). You'd have a lot better "chance" with another 2-3 though."

So basically, if a leader owns two of any of the above traits and has the early game commerce to pursue founding religions, there's a high chance for them to turn on the culture slider down the road. This is why we see Gandhi, Hatshepsut, Willem, Huayna Capac, Louis, and the like pursue cultural victory so vehemently, because they are double culture traiters who tend to found religions and turn on the slider while we see leaders like Pacal go space more often because they only have 1 culture trait and often can't found the requisite number of holy cities to make up the difference. This also has the hilarious knockoff effect where we've seen leaders like gilgamesh and kublai khan get huge and inexplicably turn on the culture slider 5% off domination because they are 1 culture traiters and have conquered the number of holy cities to trigger going into the first stage of cultural victory.

As a result, If you have rammy to win, culture is a good call. Hammurabi despite having a culture research flavor and having a love for wonders will almost never go cultural victory because he has no culture traits , and under aggressive ai, needs 7 holy cities for even the very low low chance of considering culture. He'll almost always go space as a result.

 

[AcaMetis]

Strange that the AIs are coded to consider the non-state religion spread aspect of Theocracy when they actually completely ignore that aspect. Even players can spread non-state religions to Theocracy AIs by gifting them Missionaries instead of using them directly - the player can't spread non-state religions, but the AI can.

 

[Saxo Grammaticus]

Btw. how do the playoffs work? Top 2 from each game proceeds to finals, rest are out for good?

Yes.

It has to value culture at 10 in order to pursue cultural victory, it starts at -3 because of Aggressive AI, gets 4 points for any of the traits Spiritual, Creative, Philosophical, Financial, and Industrious, 1 point for every holy city, and a random value from 0-6 depending on the location of its capital city. There's also another -3 I can't identify related to the AI's favorite civic - my best guess is Theocracy penalizes them another -3 (It's coded as isNoNonStateReligionSpread, which would suggest Theocracy).
So with only one Culture-valuing traits, you'd always need at least 3 holy cities to pursue cultural victory (-3+4+3+6 = 10). You'd have a lot better "chance" with another 2-3 though."

Thanks for sharing. I find it fascinating for a few reasons. First, this implies that with double traits, a roll of 5 or 6 on location could predispose an AI to Cultural from the very beginning. Second, anecdotally it's almost as though you can see the effect of an additional holy city, which makes sense as sometimes that pushes the AI over the victory threshold. Third, are preferences for founding/spreading and building religious infrastructure independent in whole or part of whether an AI is pursuing cultural victory?

 

[Slashninja1467]

Yes.


Thanks for sharing. I find it fascinating for a few reasons. First, this implies that with double traits, a roll of 5 or 6 on location could predispose an AI to Cultural from the very beginning. Second, anecdotally it's almost as though you can see the effect of an additional holy city, which makes sense as sometimes that pushes the AI over the victory threshold. Third, are preferences for founding/spreading and building religious infrastructure independent in whole or part of whether an AI is pursuing cultural victory?

Correct, an ai can indeed start the beginning of the game at culture 1, which is what i suspect happened with say Willem season 6. It's why we see so many ais unusually soft or go on triple missionary queues early on to spread nonstate religions.

Seeing ais spread nonstate religions to all their cities is proof of a ai moving on to culture 1 and preparing for a culture win.

 

[Fippy]

So Monty isn't that unlikely for a culture attempt?
We saw Kublai try in a championship game, his CRE and Monty's SPI give the same value?

 

[ManiaMuse89]

So Monty isn't that unlikely for a culture attempt?
We saw Kublai try in a championship game, his CRE and Monty's SPI give the same value?

In my tests for the first round game Monty actually turned on the culture slider a couple of times late in the game.

Both of those games he ended up winning by domination because he just can't help himself from declaring yer another war but one of them was quite close to being a Monty cultural victory.

 

[danjuno]

1st: Monte actually has a good start, including metals at a safe distance to discourage shenanigans. And wouldn't a Monty championship be just fantastic?
2nd: Could be Hammy or Churchill, but Hammy is further away from the low PWs, so Hammy it is.
FTD: Ramesses. Starts next to Monte.

Also this gem from the responses:

Chat is going to enjoy watching Sullla lose his mind when Churchill makes the championship for the second straight season.

I do wonder what Sulla's Joker laugh sounds like.

 

[Thrasybulos]

A 4v2 Good vs Evil is absolutely brutal for team Evil, far more so than the other way round. Probably owing to the fact that most low peaceweights can plot at Pleased, while most high peaceweights can't. In a low peaceweight-heavy game, the low peaceweights will fight each other, allowing the high peaceweights to sometimes slip through the cracks. While in a high peaceweight-heavy game, the dogpiles will keep coming.
Here, Team Evil has one thing going for it: they're both on the same side of the map, so they won't have border tension with two members of team Good. Hammurabi can be very passive, so it might be 3v2 only for a while.
Now, Monty is part of Team Evil, so if Sury gets a different religion than Monty's... there could be no "Team" Evil at all.

Ramesses is by far the stronger leader in this field, and he has double Copper in his first ring, so won't be a push-over. That should be enough to make him a clear favourite. He's bound to fight Monty, and that should not be viewed as a disaster, but as an opportunity to conquer his way to a critical mass and to add a Holy City to his collection.
Except...
He has no Horses!
In fact Team Good appears to have two Horses resources to share among them. There's one extra on Team Evil's side, so unavailable for trade during a conflict.
So while Ramesses, thanks to his Copper, will be okayish very early, he'll be absolutely crippled militarily until the Modern Era: no War Chariots, no HA, no Knights, no Cuirassiers, no Cavalry.
And no Elephants, while Team Evil has them.

So what should have been a no-brainer pick becomes a bit of quandary...

 

[Saxo Grammaticus]

I am curious why you consider Ramesses not just stronger than Suryavarman but the strongest in this playoff.

As for the lack of horses, I am less concerned following Saladin's metal-free space race victory Sure, one could focus on Ramesses' lack of access to the strong Mounted line, but one could equally point to those units' lack of defensive bonuses. In other words, perhaps we can expect Ramesses to actually mount a defense with some force other than a bunch of Horse Archers!

 

[Thrasybulos]

The way I see it, an AI's performance in a given game depends on 3 factors:

1. The game composition. And while there's a bit more to it, it can be handily approximated as the peaceweight repartition. Eg, we know Mansa is good, but Mansa in a game where all the other leaders are low peaceweights is just gonna die.
2. The starting position. Sullla puts a lot of emphasis on whether the initial available ressources match that civ's starting techs or not. And that sure has its importance. A few startling fast eliminations have drawn the community's attention to the metal situation: a civ whose only metal ressource is a 2nd or 3rd ring Iron, while its neighbour has BFC Copper, is indeed in grave danger. But the more I play those games, the more it seems the most common overriding factors there are : the quantity of available land, and the neighbouring starting positions. In a nutshell, a leader who can peacefully expand to 10 cities, while bordering a boxed-in civ... will tend to do well.
3. The AI parameters. That's what sets the different AIs apart and what we'd like to actually measure.

Now, play enough games, and the first two factors should be cancelled. Trouble is, "enough" is probably quite large.
One other caveat: there are more low peaceweight AIs in the game than high peaceweights (feature, not a bug: high peaceweights are the natural allies of the human player, the game is balanced around having a human player). So AI Survivor tournament rules will naturally favour low peaceweight AIs.

Alternate histories eliminate the 4th, unlisted factor: dumb luck. But factors 1&2 remained unchanged.

So that's why I decided to try and take a shortcut: I replay each game, but shuffling the AIs around each time. Doesn't help with factor 1, but essentially eliminates factor 2, while also providing information about the relative strengths of the starting positions on the maps (which we could guess at, but we'll have hard data to complement the guesses).

The playoffs games will use the same maps, but with different AIs (those who performed best in the Opening Round games - some are the same as what happened in the live competition, but only some).
And I've dropped the Wilcard game, replacing it with an 8-game "Wildcard League" (keeps the player count constant, and allows the eliminated civs to figure in more games).

Now, owing to the afore-mentionned imbalance between low and high peaceweight AIs, the Opening Round games favour low peaceweights, and that translates into the playoffs.
But an interesting thing about my "Wildcard League" is that it reverses that imbalance: high peaceweights outnumber low peaceweights there. I'm not done with it yet, but I've played two games (with 20 runs each) so far with a 4 vs 2 advantage for the high peaceweights... and let me tell you it was ugly for the low peaceweights.

Now, let's have a look at Season 4, game 8.
That game had a 4v3 advantage for the high peaceweights, with Hammurabi and Ramesses in their ranks.
Sullla hasn't done Alternate Histories for that one, but I have.

So Hannibal performed the best there, with Stalin, Pericles, and Ramesses not far behind as far as winning was concerned.

Now, my replays of that game, shuffle edition, have show that the best starting positions on that map were Hannibal's, Stalin's, and Pericles'. And all three performed accordingly in the AH. Ramesses still got 4 wins in spite of not being favoured by the map.
Now, here are the results where the positions are shuffled:

With 9 wins, Ramesses crushed the competition (spoiler : so far, I haven't had a game where a leader got more than 9 wins).
And you'll note that he's seriously outperformed Hammurabi.

So that's why I think that Ramesses is the best leader in that field.

But out of those 9 wins, only one was achieved with a limited core. He needs to expand, and that means attacking, not merely defending well. And the lack of Mounted units means a serious lack of offensive punch.

So I think that Ramesses is notch or two above the field... but is that enough to pull a win with essentially one arm tied behind his back?

 

[Keler]

View attachment 665285

What are Elo,Score and Score Sulla, Perf., Elo new in the table? How do you calculate that? This table looks very neat.

So AI Survivor tournament rules will naturally favour low peaceweight AIs.

Even under equal numbers low peaceweight leaders are so passive and often die, they dont help their high pw allies. They just watch them die one by one. In fact if religion spread went right, shaka&stalin could almost gonna win 2v4 that game.

So that's why I decided to try and take a shortcut: I replay each game, but shuffling the AIs around each time. Doesn't help with factor 1, but essentially eliminates factor 2, while also providing information about the relative strengths of the starting positions on the maps (which we could guess at, but we'll have hard data to complement the guesses).

No matter what pangea map you select, you see some starting positions are just hopeless and whoever you place them there with whatever neighbours they have,a civ starting that exact place won't ever win. Random map scripts always create total loser starts. This is why copy paste maps makes more sense in ranking. Afterall,neighbour randomness and AI research/production choice randomness always there. You shuffle around starting positions in a championship game. That should rank leaders better. Otherwise this becomes starting capitals ranking of a map.

 

[Kaitzilla]

Playoffs continue Friday, June 30th

Aww, another week at least.

 

[Thrasybulos]

What are Elo,Score and Score Sulla, Perf., Elo new in the table? How do you calculate that? This table looks very neat.

I'll explain in detail when I publish my results (if I keep up with the current rate, that should be in about a month's time).
But in a nutshell, I don't use Sullla's Power Rating to rank the AIs but an Elo rating instead. Mainly to not disproportionally penalize (or reward) AIs who don't play the same number of games, and also to try and account for the strength of the competition.
I do keep track of Sullla's Power Rating for comparison purposes (that's the "score (Sullla)" value).

Even under equal numbers low peaceweight leaders are so passive and often die, they dont help their high pw allies. They just watch them die one by one. In fact if religion spread went right, shaka&stalin could almost gonna win 2v4 that game.

It depends on the AIs. In this game, I expect Churchill to launch himself into the fray at some point. While Hammurabi could indeed be happy to remain in his corner and try and tech to space while the rest of the world is at war.
That's why I believe the low peaceweights have drawn an ideal setup, with border tension with only two AIs, and not the 4 of them. So Team Evil is not 100% dead in this game... only 90%.

No matter what pangea map you select, you see some starting positions are just hopeless and whoever you place them there with whatever neighbours they have,a civ starting that exact place won't ever win. Random map scripts always create total loser starts. This is why copy paste maps makes more sense in ranking. Afterall,neighbour randomness and AI research/production choice randomness always there. You shuffle around starting positions in a championship game. That should rank leaders better. Otherwise this becomes starting capitals ranking of a map.

That would be essentially correct... except playing 100s of games on the same map would bore me into dropping the project very quickly.
I'm already sick to death of that Championship map, and I'm not going to use it.

And since every AI gets to play from those overpowered or underpowered starts, that's not an issue. On the contrary, it allows to see how they try and deal with the situation, and they still get different results there.
A good AI will win every game from an OP start while others will only win some.
Conversely, a good AI might pull the odd win from a really crappy start, while other AIs will die every time.

Also, while the primary objective is to rank the AIs, the secondary objective is to provide data on the maps used in AI Survivor. Can't do that if I don't play those maps.