1000 AI Game 4.15.2 4UC+Supply Rework Stats

[L. Vern]

Hello folks,
the next installment of the "state of the AI balance" posts has arrived

These games were run both with 4UC as well as the upcoming supply rework, which generally reduces the quantity of units available at all stages of the game (all other settings unchanged from previous runs: 8 players, emperor difficulty, Communitu map, no events). Significantly more games this time as turn times are way faster and games run almost twice as fast on average!

Recent balance changes, AI improvements, and obviously the supply rework itself really shaked things up compared to 4.5 4UC , so let's get into it and see how things changed!

Spoiler Victory Types :

Great job by @axatin with the new space race code! AIs went from 430.8 ± 31.4 turns to win SV in 4.5 to 403.7 ± 27.7 in 4.15

Spoiler Civ Winrates :

tiles and cities columns refer to average percentage ownership per game

Civ	Games​	Winrate​	Culture Victories​	Diplomatic Victories​	Domination Victories​	Science Victories​	Time Victories​	Avg Pct Tiles Owned​	Avg Pct Cities Owned​	Avg Score​
The Inca	191​	0.351​	5​	10​	6​	46​	0​	0.303​	0.31​	4429​
India	198​	0.288​	33​	0​	0​	24​	0​	0.206​	0.202​	2961​
Poland	188​	0.282​	38​	3​	0​	12​	0​	0.202​	0.19​	3263​
Siam	179​	0.279​	7​	30​	1​	12​	0​	0.219​	0.205​	3451​
Greece	202​	0.252​	7​	31​	0​	13​	0​	0.254​	0.24​	3285​
Egypt	172​	0.244​	36​	0​	0​	6​	0​	0.186​	0.183​	2503​
Germany	191​	0.215​	14​	14​	1​	12​	0​	0.209​	0.198​	3177​
Austria	184​	0.212​	9​	25​	0​	5​	0​	0.173​	0.158​	2851​
The Shoshone	190​	0.2​	31​	1​	0​	6​	0​	0.214​	0.199​	2917​
Babylon	200​	0.185​	3​	0​	0​	34​	0​	0.171​	0.174​	2511​
Byzantium	170​	0.182​	8​	9​	0​	14​	0​	0.191​	0.183​	2730​
China	196​	0.179​	25​	0​	0​	10​	0​	0.191​	0.187​	2586​
Morocco	185​	0.173​	9​	20​	0​	3​	0​	0.173​	0.161​	2617​
Polynesia	205​	0.161​	9​	8​	0​	16​	0​	0.198​	0.193​	2363​
Korea	177​	0.153​	4​	0​	0​	23​	0​	0.137​	0.147​	2115​
Brazil	193​	0.15​	18​	0​	0​	11​	0​	0.168​	0.161​	2340​
The Celts	181​	0.144​	3​	6​	0​	17​	0​	0.203​	0.198​	2412​
Indonesia	188​	0.122​	7​	2​	0​	14​	0​	0.22​	0.218​	2698​
Carthage	188​	0.117​	3​	8​	1​	10​	0​	0.2​	0.198​	2939​
The Netherlands	179​	0.117​	7​	9​	0​	5​	0​	0.187​	0.183​	2567​
The Maya	180​	0.111​	5​	4​	0​	11​	0​	0.159​	0.164​	2281​
Ethiopia	185​	0.108​	6​	0​	0​	14​	0​	0.138​	0.143​	1862​
Arabia	190​	0.105​	15​	0​	0​	5​	0​	0.148​	0.148​	1981​
Russia	206​	0.092​	0​	3​	1​	15​	0​	0.237​	0.236​	2932​
Sweden	163​	0.086​	0​	5​	5​	4​	0​	0.298​	0.309​	3167​
Songhai	190​	0.084​	1​	4​	5​	6​	0​	0.295​	0.302​	3510​
England	191​	0.084​	0​	15​	0​	1​	0​	0.217​	0.228​	2871​
Portugal	193​	0.083​	1​	6​	0​	9​	0​	0.164​	0.161​	2485​
The Ottomans	181​	0.072​	1​	2​	0​	10​	0​	0.175​	0.179​	2586​
The Iroquois	174​	0.069​	2​	6​	2​	2​	0​	0.245​	0.254​	2944​
France	177​	0.068​	5​	1​	3​	3​	0​	0.248​	0.255​	2664​
The Zulus	198​	0.061​	1​	0​	6​	4​	1​	0.253​	0.262​	2599​
America	165​	0.055​	1​	3​	3​	2​	0​	0.254​	0.253​	2705​
Japan	187​	0.053​	4​	1​	4​	1​	0​	0.253​	0.263​	3052​
Assyria	187​	0.048​	1​	1​	1​	6​	0​	0.192​	0.204​	2081​
The Huns	183​	0.038​	0​	0​	5​	2​	0​	0.259​	0.266​	2673​
Rome	182​	0.033​	0​	1​	2​	3​	0​	0.207​	0.22​	2249​
Spain	185​	0.022​	1​	0​	3​	0​	0​	0.222​	0.22​	1904​
The Aztecs	183​	0.022​	0​	3​	0​	1​	0​	0.21​	0.219​	1810​
Denmark	211​	0.019​	0​	1​	1​	2​	0​	0.206​	0.216​	2025​
Persia	171​	0.018​	3​	0​	0​	0​	0​	0.179​	0.191​	1819​
Mongolia	205​	0.01​	0​	1​	0​	1​	0​	0.171​	0.179​	1603​
Venice	180​	0.006​	0​	1​	0​	0​	0​	0.13​	0.126​	1568​

• Removing the warmonger trait for Shoshone did wonders for him, going from bottom tier to among the best AI contenders
• Incas got a mountain spring water infusion or something They were always fairly scienced focused in previous results and SV is better now, though it obviously can't be the only reason or all science focused civs would be doing significantly better
• Wonder what happened to Venice, they were always a mid-high tier civ among AI contenders but now are really struggling
• Supply rework looks like it hit offense harder than defense, warmonger civs appear to be in a worse spot than last patch I ran these analyses. This has also been my personal experience playing with the supply rework, feels much easier to defend AI aggression now.

I'm looking forward to reading your guys' thoughts on what changes were the most surprising, significant, and why you think they went up or down in these stats

Spoiler Technology Research Times :

Spoiler Era Progression Times and Attainment :

Spoiler Policy Adoption and Winrates :

Spoiler Yield Sources by Era :

Cities
Instant Yields
Handicap Yields
Misc.

 

[L. Vern]

Spoiler Largest Instant/Handicap Yield Sources Breakdown :

Spoiler Religion Attainment Times :

Spoiler Pantheons :

Spoiler Founders Beliefs :

Spoiler Enhancer Beliefs + Follower Beliefs Chosen at Enhance Time :

Spoiler Reformation Beliefs :

 

[Agiwan]

Thank you very much for all the data!

I'm thinking of something: it might be good to try both a continental map like the Communitu map and a Pangea map like Pangea. Perhaps a warmonger would do much better on the Pangea map, which means the problem is not with the warmongering civilizations themselves but with the naval AI in general.

What do you think ?

 

[Bysonte]

Continental invasions are surely harder to pull off, expecially for the AI.
Pangea Communitas map when?
I would love to see it happen, unless I'm missing some start-up settings...

 

[L. Vern]

Thank you very much for all the data!

I'm thinking of something: it might be good to try both a continental map like the Communitu map and a Pangea map like Pangea. Perhaps a warmonger would do much better on the Pangea map, which means the problem is not with the warmongering civilizations themselves but with the naval AI in general.

What do you think ?

I think it's a very interesting suggestion, and a switch away from communitu for these runs is something I have been considering for a while (I have long since switched my go-to for personal play to tectonic).

I've been extremely unimpressed with and even sometimes confused at the low quality communitu map generations. In recent patches they have been frequently cramped and lopsided, with 7/1 and 6/2 splits roughly matching the gens with 4 civs on each continent.

Good ole Pangea itself is probably the most popular Pangea-style map being played in VP so will probably go with that one for such a run.

How would people feel about a map pool for these analyses instead of a single map? Ex. randomly choose between Pangea, continents, and archipelago each game rather than just one map

 

[Agiwan]

I think it's a very interesting suggestion, and a switch away from communitu for these runs is something I have been considering for a while (I have long since switched my go-to for personal play to tectonic).

I've been extremely unimpressed with and even sometimes confused at the low quality communitu map generations. In recent patches they have been frequently cramped and lopsided, with 7/1 and 6/2 splits roughly matching the gens with 4 civs on each continent.

Good ole Pangea itself is probably the most popular Pangea-style map being played in VP so will probably go with that one for such a run.

How would people feel about a map pool for these analyses instead of a single map? Ex. randomly choose between Pangea, continents, and archipelago each game rather than just one map

Would love it !

I tested the Terra-oval 1.3 map (Link) with 8 players (ai only) 4 times on King or Emperor difficulty and it ended in a domination victory 4 times/4, but I had additional modmods (Even more resources and adjustments from @hokath) and science/time victory desactivated so I guess it's not relevant here.

 

[Agiwan]

Continental invasions are surely harder to pull off, expecially for the AI.
Pangea Communitas map when?
I would love to see it happen, unless I'm missing some start-up settings...

One of the issues I see on naval AI is just the lack of units. Ai should build more boats in general.

 

[Nightmare Dusk]

I definitely think a few different maps played on would greatly benefit on giving relevant data to base stuff on. Could do a poll as well to see which maps people actually play on a bunch and based of those results play on the most popular ones (or just the one map just one utterly dominates the poll)

 

[azum4roll]

Turn 31 Classical is crazy.

What's the misc culture that only appears significantly in Information Era?

Handicap food from GP birth is HUGE.

Pantheon and founder times are very late - AI really needs to stop skipping shrines. Even if you give up on founding, building a shrine for earlier pantheon is still very worth it.
This is supported by the larger win rate of Tutelary Gods and Ancestor Worship. They aren't particularly strong, but are commonly picked first by AI. Earlier pantheon = easier win.

The graph labelled "Enhancer Beliefs" is actually for follower beliefs. Weird that cooperation is picked so much - it's really bad.

I think it's a very interesting suggestion, and a switch away from communitu for these runs is something I have been considering for a while (I have long since switched my go-to for personal play to tectonic).

I've been extremely unimpressed with and even sometimes confused at the low quality communitu map generations. In recent patches they have been frequently cramped and lopsided, with 7/1 and 6/2 splits roughly matching the gens with 4 civs on each continent.

Good ole Pangea itself is probably the most popular Pangea-style map being played in VP so will probably go with that one for such a run.

How would people feel about a map pool for these analyses instead of a single map? Ex. randomly choose between Pangea, continents, and archipelago each game rather than just one map

OVAL is the most balanced Pangaea-type map. Archipelago just makes everyone suffer.

VP is still balanced around having 2 ocean rifts separating 2 groups of landmasses though.

Communitu_79a rifts are mostly random - one starts at the lowest elevation x-coordinate and the other starts roughly at the x-coordinate that divides the land in half. But the rifts try to respect the land shape (which is random) when deciding whether to go left or right, to minimize the amount of removed land. Plus there's no guarantee the habitable landmasses of each side is the same size anyway.

 

[Circe]

Continental invasions are surely harder to pull off, expecially for the AI.
Pangea Communitas map when?
I would love to see it happen, unless I'm missing some start-up settings...

Disabling ocean rifts will make pangaea-like maps fairly common. Raise sea level too, unless you want significantly more land. Natural landmass divisions can still occur sometimes, but they might be separated only by shallow coasts allowing for earlier/easier contact.
You can fiddle with the attenuation settings in the mapscript's config file if you really want to force all land to the center, but it won't look natural.

 

[L. Vern]

Turn 31 Classical is crazy.

Yep I agree that is quite the outlier - I suppose with 8000 civ-game instances one of them is bound to pick up a ton of science ruins or something

What's the misc culture that only appears significantly in Information Era?

Sadly I don't know exactly - that number is obtained from the logs by subtracting the city, instant, and handicap yields (which are all available from the logs) from the total yields (which are available in the logs). From what I recall things like trade, vassal income, city state income, golden age bonus (the extra culture from golden age for example counts as its own thing for some reason that goes into this category), as well as traits and policies that give empire-wide yields fall into the "misc." bucket.

Pantheon and founder times are very late - AI really needs to stop skipping shrines. Even if you give up on founding, building a shrine for earlier pantheon is still very worth it.
This is supported by the larger win rate of Tutelary Gods and Ancestor Worship. They aren't particularly strong, but are commonly picked first by AI. Earlier pantheon = easier win.

Yep, it's even a few turns slower than the last patch I did these runs across the board. I agree that they pantheon and found a bit late, it's not too difficult to pick up a religion as a human player without too much investment.
You also bring up a very interesting point here which I haven't even considered, which is that the most picked pantheons are an indicator of being more common and therefore the civs that get those got their pantheon first. Warmongers really suffering though, those god of war bars look awful lol

The graph labelled "Enhancer Beliefs" is actually for follower beliefs

Yea I know these graphs have been technically incorrect since I've started making them - problem is in the logs the only available info is "picked the first time" and "picked the second" time so that's how I'm grouping them in the graphs. It's probably possible to pull the type of belief it is out of the database and merge it in to separate the two in the "enhancer" section but it's a lot of work and I'm lazy; also it's kind of interesting to see how commonly picked and how well certain beliefs perform depending on what phase they're picked

 

[azum4roll]

Yea I know these graphs have been technically incorrect since I've started making them - problem is in the logs the only available info is "picked the first time" and "picked the second" time so that's how I'm grouping them in the graphs. It's probably possible to pull the type of belief it is out of the database and merge it in to separate the two in the "enhancer" section but it's a lot of work and I'm lazy; also it's kind of interesting to see how commonly picked and how well certain beliefs perform depending on what phase they're picked

The problem is none of them are enhancers.

 

[L. Vern]

The problem is none of them are enhancers.

Wow I feel quite stupid, first I didn't understand what the problem was, then I saw I had a very simple off by one error in my parsing code
I've updated the enhancers graph, should have everything now. Thanks for bringing this to my attention

Also updated the description to be "Enhancer Beliefs + Follower Beliefs Chosen at Enhance Time" to better describe the data contained therein

 

[azum4roll]

So AI likes to pick Coorperation on enhance only, but not on found.

 

[L. Vern]

So I'm just noticing that Inca have the highest average cities and highest average plots owned per game (by far highest average score too lol)... I think they might actually be the best warmonger.

Interestingly, I've actually seen them lose a game because of this - they were at future tech but were on a 3 front war and I guess were trying for a DomV so didn't build any spaceship parts at all - let Germany take an SV right from under their nose

 

[axatin]

Interestingly, I've actually seen them lose a game because of this - they were at future tech but were on a 3 front war and I guess were trying for a DomV so didn't build any spaceship parts at all - let Germany take an SV right from under their nose

hmm, we need to have a look at the grand strategy logic then. At some point they should have noticed that they are closer to SV than to DomV.

 

[azum4roll]

A human would go for both, but humans also don't need to build units.

 

[hokath]

Interesting that Statecraft and Industry are less popular with AI, but have better winrates than other trees. Even AI is sleeping on Industry!

It would be interesting to have the statistic for: "proportion of Diplomacy victories where the winner adopted Statecraft".

 

[randomnub]

I think it's a very interesting suggestion, and a switch away from communitu for these runs is something I have been considering for a while (I have long since switched my go-to for personal play to tectonic).

I've been extremely unimpressed with and even sometimes confused at the low quality communitu map generations. In recent patches they have been frequently cramped and lopsided, with 7/1 and 6/2 splits roughly matching the gens with 4 civs on each continent.

Good ole Pangea itself is probably the most popular Pangea-style map being played in VP so will probably go with that one for such a run.

How would people feel about a map pool for these analyses instead of a single map? Ex. randomly choose between Pangea, continents, and archipelago each game rather than just one map

I've been playing vanilla pangea and continents only for years, as they feel like the most balanced map scripts for the AI.

I eventually grew bored of it and in the past months gave a try to communitu and tectonics, mostly standard settings and it felt a bit like cheating, I think the difficulty goes down a notch by picking those map scripts no matter how you cook the settings: they might make for prettier looking maps but all the rough terrain, mountain ranges, rivers benefit the human more, and warmonger AIs less. My default promotion path for melee in the first 3 eras has been shock into woodsman into amphibious, and in later eras I don't bother with 2 moves units anyway. No surprise no AI even came close to contest a peaceful victory of mine even on deity, as it takes ages for them to move troops around.

Those custom map scripts also give me more room to peaceful settle some decent spots early (making tundra and desert smaller in communitu, or putting more resources into them in tectonics) and later (plenty of small island chains worth plop a pioneer for a pop 6-10 city), they don't reward early aggression and conquest. Tectonics is simply huge at standard size, luckly it can be shrinked but even with the max 85% reduction I find myself with plenty of space to peacefully settle.

As much as I enjoy the random components of map generation (the early exploration, the ability to settle good cities even in mid-late game) I'd stick to vanilla map scripts when it comes to balance AIs. Then it's up to the player deciding if they want to have a funnier, easier or harder game during game setup.

 

[azum4roll]

making tundra and desert smaller in communitu, or putting more resources into them in tectonics

Both are false...
Standard maps are completely random in a sense that there can sometimes be 1% desert and sometimes half of the map is desert. Same for tundra/snow/marsh. It's more consistent in Communitu_79a.
I can't speak for Tectonics but the map uses standard resource placement IIRC.

plenty of small island chains worth plop a pioneer for a pop 6-10 city

This is a feature. How else are you ever going to use pioneers/colonists?

There are numbers that can be adjusted to make fewer hills. Rivers are more complicated.