AI League (yet another AI Survivor spinoff)

[Thrasybulos]

When I launched my AI Jumbled Rumble last year, it was to provide :
- An alternate (better ?) ranking system for the AI leaders
- An assessment of the balance of the maps used in AI Survivor (which would help explain the Alternate Histories results, and hopefully help us "read the maps" better).

After two "seasons" of Jumbled Rumble, I would say the second objective is largely fulfilled, but I'm not satisfied as far as the first, primary objective is concerned.
While some interesting insight was provided about the importance of the diplomatic context (beyond the obvious high peaceweight leader in a low peaceweight field = bad), playing 20 instances of a specific context leads to a distortion effect which runs contrary to an attempt at establishing an overall ranking for the AIs.

Chess was an obvious inspiration for the Elo rnaking, so Chess will help here too : I'm going to borrow from the Open (or "Swiss") tournament system.

Each Tournament will feature 8 "arenas".
The "top" 4 arenas will host 6 players.
The "bottom" 4 arenas will host 7 players.

A tournament will be played over a fixed number of rounds.
I'm aiming at 10 rounds, but depending on how things go for the first tournament, that might be extended a bit (to 12 ?).

For the inital round, each leader will be randomly assigned to an arena.
For subsequent tournament, I might introduce a seeding system, but at this point, I'm inclined not to.
After that, the top performers will be assigned to the top arenas, the worst performers to the bottom arenas.
For instance, for round 2, Arena 1 will host 6 of the 8 winners from round 1, while Arena 8 will have leaders which were all eliminated in round 1.

Assignment of a leader to an arena for each round will be a function of that leader's running score.
Ties will be randomly broken. As the tournamnent progresses, there should be fewer and fewer of those.

At the end of the last round, the leader with the highest score will be that tournament's winner, a purely honorific title.

That should fix many of the shortcomings of the previous system :
- Every leader will play the same number of games.
- We'll get to see a wide variety of match-ups (in the initial rounds).
- But we'll also have leaders compete with leaders of a similar strenght (in the later rounds).
(It should be noted though that this last point means that compared with playing purely random games, this will lead the better leaders to somewhat underperform, and the worst leaders to somewhat overperform : in an all-star game, there can be only one winner, and conversely, in an all-loser game, one has to win.)

Each arena will be assigned a map randomly drawn from Sullla's AI Survivor series, and that map will stay the same for that arena for the duration of the tournament.
In that way, what started with the Rumbled Jumble will continue here.
Leaders assigned to an arena will then be assigned a random starting position on the map, with no repeats : if a leader played on that map previously, it will be assigned a different starting position.

The "League" (inspiration has failed me, there) will be a succession of tournaments ("seasons").


Game parameters & setup :

• The games follow Sullla's parameters : Aggressive AI, No Tech Trading, No Events, No Vassals, Deity Difficulty but without bonus starting techs for the AI.
• As was the case for Jumbled Rumble, I have also disabled the Diplomatic Victory condition.
• Contrary to Sullla, I'm not using settled Great Spies to get access to the various graphs. Also of very minor note, I've removed the Observer civ's starting techs.
• When using an AI Survivor Wilcard game map, Raging Barbs are turned OFF.

 

[Thrasybulos]

Scoring

I will be using and keeping track of different scoring systems.

ScoreS (survival score).
The idea with Jumbled Rumble was to introduce an Elo rating for the AI players.
I'm going to keep that idea alive.
An Elo rating uses an underlying scoring system, and here I'll be using a simplified version of the one I used for Rumbled Jumble.
For a given leader, a game has three possible outcomes : win > survival > elimination.
A leader's score will be computed by comparing his outcome with that of the other leaders, one by one.
Better outcome, score 1, worse outcome, score 0, same outcome, score 0.5.
So, for a typical 6-player game which sees 3 eliminations :
The winner gets 5 points.
Both survivors get 3.5 points (1 vs each eliminated leader, 0.5 vs the other survivor, 0 vs the winner).
The 3 eliminated leaders gets 1 point each (0 vs the winner and the survivors, 0.5 vs each other survivor).

In RJ I had a finer-grained system in place, so that the earliest elimination would lead to a lower score, and a contender for the win would get a better score than a one-city "zombie" civ.
While it felt more satisfying, in the end the impact is minimal and it adds a lot of complexity. So let's KISS it good-bye.

Now, one issue which has become apparent over the games, is that it somewhat overvalues survival.
Let's get back to our typical 6-player game with a 5 / 3.5 / 1 result.
An AI winning a game and being eliminated in the next game would score 6 points.
An AI not winning but surviving both games would score 7 points.
Thus, a better score, while it is very debatable that it's a better performance...

There had been at the time some discussion about how to devise a scoring system which would adequately reflect an AI's in-game performance.
I pretty much came to the conclusion that it would be a fool's errand, but wondered whether scoring "performance" shouldn't be dropped altogether, in favour of scoring wins only.

ScoreW (wins score).
The game winner gets a win (1 point) vs every other leader.
And that's it.

FTD and runner-up are considered the same : they've lost.
This conveniently solves the issue about how to rate survival vs winning : survival is not a consideration.
Under that system, a higher score is not an indication of "doing better" (whatever that means) but of "winning more" (meaning is clear).

Of course, this also comes with issues.
It is debatable whether totally dismissing any notion of performance beyond winning / losing should lead to a better ranking system.
And game composition gets disregarded for the losing AIs : losing to HC in a game featuring Mansa, Pacal, Justinian is the same as losing in a game with Frederick, Monty, Roosevelt. It's a loss to HC, period.

Score (combined score).
So... we have a scoring system which overvalues survival, and one which undervalues it : why not simply combine both scoring systems ?
Score = ScoreS + ScoreW

In this system, for the standard 6-player game, winning once and getting eliminated 3 times would net 10 + 1x3 = 13 points.
Surviving 4 times would net 3.5x4 = 14 points.
That's probably fairer.

So this Score is what I'll be using to rank the AIs in each tournament.

Elo ratings.

I'll also provide elo ratings for each of these scores.
But that'll come later, I'll need to set up the tool for it first.
And anyway, since the elo ratings will be updated after each tournament and not after each game, it will take a few tournaments for the elo rankings to differ from their underlying score rankings.

PWR (Sullla's Power Rating).

I'll also be keeping track of Sullla's power rating for each AI (5 points for a win, 2 points for finishing second, 1 pt / kill).
It should be noted that while this is the exact system Sullla uses in AI Survivor, it's going to be different.

That's because in AI Survivor, the AIs don't play the same number of games. Scoring in the opening rounds provides the opportunity to score in later games, while non-scoring AIs don't get that opportunity.
Here, all AIs will play the exact same number of games.

NScore (normalized in-game score).

The game does provide its own scoring system, with the in-game score.
I'll keep track of that too : maybe after all it's reliable enough ?

The one adjustment I'll make is to "normalize" that score for the game winner only.
For surviving leaders, NScore = in-game score at the end of the game.
For the winner, NScore = in-game score x 500 / Victory Turn.

This is to help offset the fact that early victories (especially Culture) tend to be severely undervalued in-game.

 

[Thrasybulos]

Tournament 1

• Round 1
• Round 2
• Round 3
• Round 4
• Round 5
• Round 6
• Round 7
• Round 8
• Round 9
• Round 10

Tournament 2

• Round 1
• Round 2
• Round 3

 

[Thrasybulos]

Tournament 1
Since I've already played the AI Survivor seasons 3 & 4 maps extensively, I gave them a rest for now.
And with Seasons 1 and 2 maps not available yet to me, I drew from Seasons 5, 6, 7.

So, here are the maps which were selected.

Arena 1 : Season 5, Playoffs, Game 1
Arena 2 : Season 5, Opening Round, Game 1
Arena 3 : Season 7, Opening Round, Game 3
Arena 4 : Season 6, Opening Round, Game 5
Arena 5 : Season 5, Wildcard Game
Arena 6 : Season 5, Opening Round, Game 8
Arena 7 : Season 6, Opening Round, Game 8
Arena 8 : Season 6, Opening Round, Game 6

A few things to mention :

• I've cleared the ice from the sea in a consistent manner across all maps. This has no impact, except for a few maps where it creates viable fishing villages spots which weren't there for Sulla's game (especially true for season 5 maps, or Season 3 Opening Game 1).
• Fixed colours are assigned to starting positions, not leaders. It gets some using to, but it'll make discussing positions easier, ditto should I develop tools to generate the files instead of editing them manually.
• I've re-ordered the starting positions in a consistent manner across the maps : top left first, then clockwise.
  It helps reading and should have minimal impact... except when it does. The very first game I played was on the Season 5 map where Alex lost the settler race to Pacal, and his access to metal by the same token. Pacal, as a pool leader, had the turn order advantage. With my reordering, this gets reversed : the exact same settler race happened, with the opposite result. So that starting position is made a little less weak as a consequence.

Round 1

Spoiler Arena 8 :

Spoiler Arena 7 :

Spoiler Arena 6 :

Spoiler Arena 5 :

Spoiler Arena 4 :

Spoiler Arena 3 :

Spoiler Arena 2 :

Spoiler Arena 1 :

Spoiler Rankings :

Note : The replays and worldbuilder files are available should you guys want them.
But there's no point uploading them if no one's going to use them, so I'm not going to clutter the forums with them unless someone wants to have a look.

 

[Thrasybulos]

Tournament 1, Round 2

Spoiler Arena 8 :

Spoiler Arena 7 :

Spoiler Arena 6 :

Spoiler Arena 5 :

Spoiler Arena 4 :

Spoiler Arena 3 :

Spoiler Arena 2 :

Spoiler Arena 1 :

Spoiler Rankings :

 

[Thrasybulos]

Tournament 1, Round 3

Note : While copy-pasting, I messed up a formula, granting an extra point to the winners of arenas 1-4 for rounds 2 and 3.
It's fixed now, but the values I used for attributing players to an arena were slightly off for rounds 3 and 4.

Spoiler Arena 8 :

Spoiler Arena 7 :

Spoiler Arena 6 :

Spoiler Arena 5 :

Spoiler Arena 4 :

Spoiler Arena 3 :

Spoiler Arena 2 :

Spoiler Arena 1 :

Spoiler Rankings :

 

[Thrasybulos]

Tournament 1, Round 4

Spoiler Arena 8 :

Spoiler Arena 7 :

Spoiler Arena 6 :

Spoiler Arena 5 :

Spoiler Arena 4 :

Spoiler Arena 3 :

Spoiler Arena 2 :

Spoiler Arena 1 :

Spoiler Rankings :

 

[Thrasybulos]

Tournament 1, Round 5

Spoiler Arena 8 :

Spoiler Arena 7 :

Spoiler Arena 6 :

Spoiler Arena 5 :

Spoiler Arena 4 :

Spoiler Arena 3 :

Spoiler Arena 2 :

Spoiler Arena 1 :

Spoiler Rankings :

And this is where I've come so far, half-way through the first tournament (provided I settle as planned for a 10-round format).

Although there are a few unexpected names both at the top and the bottom, the most surprising fact to me is that by and large, leaders are where we'd expect them to be.
Keep in mind that contrary to the AH or their jumbled version, games are not repeated to get at a "correct" picture. Here, the "outlier effect" is in full swing as it is in the live AI Survivor event, and yet...

We'll see how that holds after 5 more rounds.

 

[NickGabben]

That's a heck of an effort, chapeau

 

[lymond]

I've read thru half so far. Not quite sure if there are different parameters to Sulla. Sally seems to be doing very well.

 

[Saxo Grammaticus]

Keep in mind that contrary to the AH or their jumbled version, games are not repeated to get at a "correct" picture. Here, the "outlier effect" is in full swing as it is in the live AI Survivor event, and yet...

Round 2, Arena 5 Ragnar with a Cultural Victory. To my knowledge, a leader who has never won a game across the tournament or the Alternate Histories. Hope that was fun to watch!

I am fairly surprised with the bottom of the rankings after five rounds. Perhaps some of those leaders will take off in the next five, but I get the impression of consistency issues between the top and bottom.

Thanks for sharing! I have no need for the replays and maps at present but appreciate the offer.

 

[Thrasybulos]

Not quite sure if there are different parameters to Sulla.

The main difference is no UN.
I'm also running a lower impact Observer civ (no Great Spies), but weirdly enough, that doesn't seem to have a really noticeable effect.

Round 2, Arena 5 Ragnar with a Cultural Victory. To my knowledge, a leader who has never won a game across the tournament or the Alternate Histories. Hope that was fun to watch!

It was the standard warmonger with a cultural trait (Financial) conquering enough Holy Cities to trigger the Cultural Victory attempt. Usually it either throws their game (they get out-teched) or just speeds up Domination with border pops, but here it struck the perfect middle : it didn't cost him his game, but his progression towards Domination was slow enough that the Cultural Victory hit.

I am fairly surprised with the bottom of the rankings after five rounds. Perhaps some of those leaders will take off in the next five, but I get the impression of consistency issues between the top and bottom.

Some leaders can luck on good starting positions and unexpected diplomatic developments and thus get better than average results.
Or they can get unlucky with their starting position and/or diplomatic environment and face a real uphill struggle.
Once I put the results into a real database (only so much you can do with Excel), that's one of the many stats I'll be able to extract (how "lucky" each leader has been with their assigned starting positions).

Anyway, a single tournament is not meant to provide an accurate ranking. It would take quite a few for that.

 

[Eauxps I. Fourgott]

Is posting this while halfway completed your way of poking me to get and send you those WorldBuilder files at last?

 

[Keler]

One issue is if lets say someone with high pw gets stuck on an arena that gets filled with low pw, they need to wait couple of rounds for some low pw push other high pw down to help out. But of course this looks fun and shuffles things around. I really liked those replay gifs. Great work there.
Since we look at 40 games so far, it needs some time for ranking to become more consistent/stable.

I believe world builder files for season 2 shared by Eauxps in your old thread. Season 1 available from Sulla s website too.

 

[Thrasybulos]

Is posting this while halfway completed your way of poking me to get and send you those WorldBuilder files at last?

Yes, of course !
No, I just wanted to wait a bit to make sure I wasn't unhappy with the way it worked, but didn't want to wait until I was finished so that it would make for a more "digestible" chunk when posted, and so that I might get some feedback if needed.

One issue is if lets say someone with high pw gets stuck on an arena that gets filled with low pw, they need to wait couple of rounds for some low pw push other high pw down to help out.

I believe world builder files for season 2 shared by Eauxps in your old thread. Season 1 available from Sulla s website too.

Thanks for the files ! Had I missed them ? I thought they were only standard saves, not WB saves ?
Anyway, doesn't matter, I have them now.
I'll probably include the season 2 maps in the pool of maps to draw from starting with Tournament 2.
For the season 1 maps, I can wait until Eauxps is done with the AH.

And yes, some AIs may get stuck in an unfavourable environment for a few games. That's why I don't expect a consistent performance over the iterations, but that shouldn't matter with enough total games played. Hopefully.

 

[Keler]

For standard save files you need them to copy paste into Saves single file above WorldBuilder, then open the game and load them and save as worldbuilder file in game. I mean season 3 game files were shared as save files and at turn 1 which you have used them?

 

[Thrasybulos]

I know that.
The issue is I can't open Sullla's saves without my game crashing.
Removing my modded files, using the GoG or Steam version, still the same.
For the season 3 maps, Saxo provided me with the wb saves.

 

[Thrasybulos]

Tournament 1, Round 6

Spoiler Arena 8 :

Another very poor showing by Willem who completely fails to expand initially. Augustus takes advantage of his weakness and starts devouring him.
Mehmed comes to the rescue, only to be immediately backstabbed by Monty with whom he had just been allied for the conquest of Washington.
That opens up the way for Mao, who has peacefully developped, to become the dominant AI and take the game.

Spoiler Arena 7 :

Another game of betrayals.
Julius and Alex start conquering Victoria, but Alex peaces out and joins Viccy to defeat Julius. And then he right away resumes hostilities against his erstwhile ally, with Stalin joining the party.
In the East, Tokugawa saves Genghis from being conquered by Frederick. The Khan shows his gratitude by peacing out and declaring on him.
But Toku is the early game's dominant AI and he's winning the 1v2 hands down.
Two issues for the Japanese leader :
a) He's founded one of early religions. But founding a religion when you won't let it spread through open borders is a sure way of becoming a religious pariah.
b) He's the research leader with twice the beakers of the next AI. As such, he's doing the Willem not-so-special by going after all the first-to bonuses on the tech tree while ignoring every single military tech.
So when Alex and Stalin, who have been getting new military toys, come knocking, the Japanese empire crumbles very fast.
Somehow Alex has managed to completely out-eco Stalin. That doesn't stop Stalin from triggering the last war against his much more advanced and much more powerful former friend.
Alex achieves Domination while his ship is on its way to Alpha Centauri.

Spoiler Arena 6 :

As the only low peaceweight leader in this field, Cathy seems doomed. But she was lucky in drawing the best starting position, next to the two death spots (red and blue, Darius and Roosy) on this map.
With Lincoln, Asoka, and WK having each founded their own religion and thus locked in religious wars, and with Joao seemingly checked out of this game (to be fair, shared peaceweight and/or religion + HR as favourite civics + cannot plot at pleased kinda tied his hands...), there seems to be a narrow window for her.
Darius, stuck on 3 cities, fights bravely but falls. Then, rather than finish off Roosevelt, Cathy joins the current religious dogpile on Lincoln and conquers most of his lands.
And then... while far from the tech lead, and with a military tech flavour, she goes Democracy, Corporation, Sci Meth, and won't even research Gunpowder, let alone Rifling. And chooses to declare on Joao, the only leader with more power than her, having just unlocked infantries ! Bye, Cathy.
Asoka pulls the slider for a weak Cultural Victory attempt, and WK, who's played a decent game so far, goes back to his trolling ways : instead of punishing his game-long enemy, he joins the war against Cathy (and goes full troll by claiming the kill credit).
With Cathy gone, Joao could still easily put an end to Asoka's attempt (he's way, way stronger, and "cautious" so he can declare). But he'd rather do nothing and lose the game, thank you very much.

Spoiler Arena 5 :

Hatty is doomed in this game, and indeed by turn 90, she's already in a 1v4 situation.
I guess she knows it and that she chooses to build missionaries instead of settlers, and to research every religious tech instead of worker techs and securing metal, so that when she inevitably gets conquered her conquerors get poisoned-pilled with her holy cities and she gets the last laugh as they fall upon one another over religious differences ?
Her plan may work, as after her early elimination, there are no less than four major religions in the world.
The other high pw leader, Zara, also gets the major dogpile treatment.
Peter, who sneak-attacked De Gaulle earlier, is called to a reckoning.
But apart from that, Hatty's plan fails : in spite of the religious differences, similar peaceweight + shared military struggle means this bunch of leaders who can't plot at pleased stay at peace (only Ragnar could, but he's the weakest).
The game turns into an uneventful spaceship race between the two major players, Brennus and De Gaulle. Brennus, who had a headstart, keeps his advantage and wins with a 3-tech lead.

Spoiler Arena 4 :

The three initial religions get founded by each of the three leaders at the cramped top of the map. Should get... interesting. Especially as each religion spreads to the leader to its south, creating three blocks. And to add to the mayhem there, Hannibal settles the map in a crazy fashion, managing, from his bottom center position, to wedge a city between Isabella and Hammurabi, at the top of the map.
Except it doesn't happen, the game unfolding along expected peaceweight lines.
Hannibal and Gilgamesh carve up Elisabeth (actually a bad deal for Gilgamesh as, for geographical reasons, Hannibal is set to benefit more ; and an especially bad deal when Hannibal lucks out and gets most of the spoils).
Sury slowly conquers Hammurabi but gets Victus Interruptus when Hannibal comes knocking (Giggles gets to finish off Hammy).
Gilgamesh is by far the tech leader and could have at any point taken control of the game, but he seems content to watch Hannibal conquer the world while he techs away.
Isabella, who had been plotting against Sury for ages without pulling the trigger, watches him die and goes back to sulk in her corner.
At this point, Hannibal is close to Domination. He's Annoyed with Isabella and could very easily crush her and take the game.
Instead, he choses to go after the much more advanced Gilgamesh he's merely Cautious with.

Yeah, that goes poorly.
Hannibal manages to somewhat stabilize after a while, but the damage is done and he sues for peace.
As Gilgamesh puts the finishing touches to his spaceship, Isabella could have thanked everyone for the ride. She decides instead to suicide into Hannibal.
The Sumerian ship gets to its destination just in time to save her.

Spoiler Arena 3 :

Ramesses, thanks to the Egyptian special - a super early shrine -, converts everyone to his religious, except for Charlie who's following his own.
As if starting in a weak spot, next to Shaka, wasn't enough, Charlie is now the world's religious pariah.
There's no surprise ending to that story: Shaka gets the spoils, Louis gets the kill.
Ramesses is the tech leader: he founds 4 more religions. I guess we can see where this is going...
Shaka is bored with this religious lovefest: he picks a fight with the next strongest guy on the block, Bismarck.
Louis takes his cue and decides he wants a piece of England for himself.
The dice! They don't work! Ramesses, with 5 Holy Cities, sheltered behind Bismarck and Churchill, has the perfect setup for a very early Cultural Victory... but he can't get to roll the trigger. So instead, he's limping to a 10-city, non-financial spaceship attempt.
The dice finally cooperate on turn 280 when he pulls the trigger. But Bismarck and Churchill are gone. And so is his religous good luck charm as he goes into Free Religion.
Sure enough, Louis comes crashing. Then Shaka.
Memphis falls with 5 turns to go. Ramesses would have won... IF HE HADN'T TURNED OFF THE SLIDER!! Seriously, this guy. In this tournament alone, this is the second game he completely throws when he had a sure win!
With no one else to fight, Shaka decides to roughen up Louis... but here's a tidbit of history they don't teach you at school. Although Shaka has a decisive tech lead, Louis has made it a priority to justify his "Sun King" title.
More precisely, the "Radiance of a Thousand Suns King".

Shaka's main stack gets vaporized in the first turn of the war.
Dejected, the Zulu has to accept his defeat science victory.

Spoiler Arena 2 :

Sully gets unlucky as his neighbours, and the rest of the world for that matter, pick up HC's religion instead of his (and to add insult to injury, it is Islam HC has chosen to found).
His bad luck continues as he teams up with Kublai to carve up Mali, and the Mongols get every city but a tundra colony.
HC plays a poor early game: he fails to expand beyond 5 cities, and then picks up a fight with Sitting Bull who, while not much bigger, is a tad stronger.
Sully has played the best early game, but Kublai's conquests put him ahead. The Ottomans try to remedy that by conquering Greece. Pericles' core cities and capital fall quickly.
And then, obviously, Kublai backstabs Suleiman who is actually a bit stronger, but whose armies are on the other side of the map.
Sully manages to get peace though, while losing a single city, and he can finish off Pericles.
At this point, SB and HC seem irrelevant, locked in a mutually destructive conflict. So the game looks like it's going to be Kublai's or Suleiman's. Probably the former, as the Mongols have better land, a better diplomacy, and an economy headstart (Kublai has been fixing his economy while Sully was fighting in Greece).
Suleiman has more power, though, so he plays that card by attacking Kublai.
But he directs his research into lalaland, while Kublai goes straight for Rifling and Assembly Line, and that's all she wrote.
Suleiman eventually gets peace (a reprieve), but the damage is done.

Spoiler Arena 1 :

Gandhi shouldn't outstay his unwelcome in this game for very long. And with Pacal next him having inherited the strongest start on this map, this game feels like it should follow a predictable scenario: Gandhi FTD, Pacal wins.
Easy peasy.
Saladin and Pacal found the first religions, with everyone adopting Pacal's. Except for Gandhi, who elects to follow Pacal's minority religion. That's the spirit: assume your unpopularity!
Qin has sent his first settler the wrong direction, and he thus fails to secure metal. Saladin pounces.
Cyrus is astonishingly the only one to come knocking on Gandhi's door. Qin is busy dying, Saladin is busy killing him, Pacal is busy building wonders, Justinian is busy... I don't know. He's apparently not even plotting against the Arab infidels.
Close race between Qin and Gandhi, but the Chinese leader is the first one out of the game.
Saladin emerges as the game leader, but he's walking on diplomatic eggs.
He gets a break when Cyrus converts to Islam. He decides to break an egg. With Justinian's face on it.
Pacal has dodged a bullet there, but he gets the message: time is running short. He pulls the trigger on the culture slider.

Spoiler Rankings :

A lot of familiar names at the top. No high peace-weight victim for the next Arena 1 game, but a possibility that Round 8 could feature three high pw leaders there.
At the bottom, Willem and Mehmed's woes continue...

 

[ManiaMuse89]

One issue is if lets say someone with high pw gets stuck on an arena that gets filled with low pw, they need to wait couple of rounds for some low pw push other high pw down to help out. But of course this looks fun and shuffles things around. I really liked those replay gifs. Great work there.
Since we look at 40 games so far, it needs some time for ranking to become more consistent/stable.

I believe world builder files for season 2 shared by Eauxps in your old thread. Season 1 available from Sulla s website too.

I sort of agree.

It is giving me nightmares of being in a chess club at my primary school which had a ladder system with promotion and relegation between two leagues. To get promoted you had to be at the top of the bottom league and then have a playoff with the bottom player in the top league. If you won you got to swap leagues but it was very easy to get immediately relegated again because you would usually end up playing another player who was similar ability in another promotion/relegation playoff. But otherwise you never got to play the better players in the top league because they never fell down the rankings enough and just meant that you were playing the same players all the time.

I think a more random draw system to make the leaders play a bigger variety of different opponents would be better. There are definitely some AI leaders who have 'bogey' AI survivor opponents or end up always working well together against the rest of the field for various reasons.

 

[Thrasybulos]

I think a more random draw system to make the leaders play a bigger variety of different opponents would be better.

Two issues with that :

• I want the "best" AIs to play together at some point so they measure up against one another. Same for the "worst" AIs.
• Running a system where there's a winning condition and "progression" is just, for me at least, way more fun than just running random games ad nauseam. Put it in other words, I want a game, not just raw data collecting.

And there's actually already plenty of variety : the first round is completely random, and then the big "jumps" that occur with each win ensure some shuffling around.
It's just at the very top and very bottom that there's less variety... and that's kinda by design.

And by the way, I had (of course ?) entertained the idea of having two "pools" / "leagues" with a promotion/relegation mechanism.
My experiment with the "wildcard league" last year steared me away from it: I don't want to create two very separate diplomatic environments, where some leaders would thrive in one and get thrashed in the other.