650 AI Game 4UC Stats / Analysis

L. Vern · May 10, 2024

Hello all,
Here's a stat dump of 650 4UC games played off of V4.5, to get as close a comparison as possible between 4UC and base VP. Same settings as previous run: communitu map, 8 player emperor.
The base version is the same, and number of games is the same, so let's take a look at what difference 4UC makes as well as stats for 4UC in general.

Spoiler Civ Winrate Changes :

Some highlights of base VP vs 4UC:

Biggest Winners

Carthage: 22 ranks, 9.8% WR
Siam: 8 ranks, 7.4% WR
India: 9 ranks, 7.1% WR
Byzantium: 11 ranks, 6.7% WR
Polynesia: 10 ranks, 5.5% WR

Biggest Losers

Ottomans: 17 ranks, 9.3% WR
Netherlands: 8 ranks, 7.5% WR
Indonesia: 14 ranks, 7% WR
Venice: 11 ranks, 6.2 WR
Aztecs: 14 ranks, 6% WR

Spoiler Winrate Quantiles :

Key takeaways:

Base VP has higher win rates in the top quantiles, meaning the top base VP civs are more powerful than top 4UC civs
Base VP has a smaller slope in bottom quantiles, meaning bottom base VP civs are less powerful than 4UC civs
4UC has a slightly tighter winrate spread than base VP
The winrates are roughly normally distributed (a perfect normal distribution would be a straight line on such a plot)

Spoiler Victory Type Comparison :

4UC games are a more consistent length on average, and the median 4UC game takes slightly fewer terns than the median base VP game.

The remaining sections are stats from 650 4UC games

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
Siam	113	0.274	8	16	0	7	0	0.167	0.154	1921
Austria	122	0.262	8	17	0	7	0	0.147	0.137	1964
Germany	110	0.245	5	19	0	3	0	0.158	0.145	1640
Korea	126	0.23	10	0	0	19	0	0.173	0.172	1910
Poland	99	0.222	14	3	0	5	0	0.155	0.147	2077
Arabia	114	0.211	20	0	0	4	0	0.144	0.139	1568
Songhai	132	0.205	1	4	14	8	0	0.408	0.418	3506
China	103	0.204	17	0	0	4	0	0.177	0.173	1773
India	124	0.202	17	0	0	8	0	0.162	0.154	1678
The Netherlands	110	0.2	3	17	0	2	0	0.187	0.177	1713
Byzantium	119	0.176	0	11	0	10	0	0.155	0.149	1775
The Maya	116	0.172	4	5	1	10	0	0.188	0.187	2045
The Inca	119	0.168	1	0	1	18	0	0.256	0.26	2917
Greece	119	0.16	1	15	0	3	0	0.193	0.182	1667
Carthage	100	0.15	0	12	0	3	0	0.223	0.227	2074
Babylon	121	0.149	9	0	0	9	0	0.164	0.161	1979
Egypt	109	0.147	13	0	0	3	0	0.157	0.15	1609
Venice	124	0.145	6	8	0	4	0	0.177	0.165	2250
Polynesia	105	0.133	1	9	0	4	0	0.2	0.202	1927
Morocco	121	0.132	1	14	0	1	0	0.127	0.117	1469
Portugal	115	0.122	2	2	0	10	0	0.192	0.192	1884
Russia	119	0.118	2	4	0	8	0	0.249	0.251	2304
The Ottomans	113	0.115	1	5	1	6	0	0.222	0.223	1946
The Zulus	114	0.114	0	0	10	2	1	0.37	0.372	2965
The Celts	129	0.101	2	5	0	6	0	0.169	0.164	1601
Brazil	103	0.097	3	3	0	4	0	0.148	0.145	1549
Sweden	126	0.095	2	0	7	3	0	0.306	0.313	2677
Japan	125	0.088	4	1	0	6	0	0.186	0.183	1952
Assyria	111	0.072	1	3	0	4	0	0.185	0.187	1518
England	114	0.07	0	7	0	0	1	0.208	0.221	2274
Indonesia	128	0.07	1	3	0	5	0	0.249	0.249	1923
Ethiopia	134	0.067	5	0	0	4	0	0.145	0.147	1345
Denmark	106	0.057	1	2	1	2	0	0.256	0.27	1952
America	127	0.055	2	3	2	0	0	0.287	0.29	2529
France	111	0.054	2	1	1	2	0	0.277	0.283	2400
Rome	117	0.051	1	3	2	0	0	0.279	0.291	2215
Persia	138	0.051	3	2	1	0	1	0.25	0.259	2058
The Iroquois	102	0.049	0	3	1	1	0	0.291	0.301	3080
The Shoshone	114	0.044	2	2	0	1	0	0.222	0.219	2059
Mongolia	106	0.038	0	0	2	2	0	0.232	0.237	1611
The Huns	111	0.036	2	0	0	1	1	0.24	0.237	1935
The Aztecs	101	0.02	1	0	1	0	0	0.259	0.267	2042
Spain	122	0.008	1	0	0	0	0	0.264	0.264	1795

Spoiler Technology Research Times :

Spoiler Era Progression Times and Attainment :

Spoiler Policy Adoption and Winrates :

Spoiler Yield Sources by Era :

L. Vern · May 10, 2024

Spoiler Largest Instant/Handicap Yield Sources Breakdown :

Spoiler World Congress Resolution Enact/Repeal Frequency :

Spoiler World Congress Resolution Proposal Frequency Across Games :

Spoiler World Congress Vote Deltas by Resolution :

Spoiler Religion Attainment Times :

Spoiler Pantheons :

Spoiler Founders Beliefs :

Spoiler Enhancer Beliefs :

Spoiler Reformation Beliefs :

msw1 · May 10, 2024

Absolutely fantastic, thank you for this!

rkkn · May 10, 2024

I still have never played 4UC, but Carthage being the biggest winner is not at all surprising. Back when I first looked at the mod I thought "wow that's OP" when I saw the Carthage UB

interesting though that the overall balance appears to be (marginally) better

rkkn · May 10, 2024

a few of the pantheons have astonishingly low win rates
springtime, purity, and spirit of the desert have a green part that is very small relative to the rest of it. And, surprisingly, god of the sea as well

CppMaster · May 10, 2024

Seems like another good argument to integrate 4UC mod

Anarcomu · May 10, 2024

I think there is a problem in the religious part :
There are 5 type of belief : Pantheon, Founder, Followers (you get 2 of them, one at founding, one at enhancing), Enhancer, Reformation
You have not made the Followers part and there is followers in both Founder an Enhancing.

Aaand I think I get it. Is this what AI pick when they found/enhance ?

pineappledan · May 10, 2024

Fewer Time victories, and more domination victories.
Slightly shorter games overall, but also eliminated sub-turn 300 wins.
Smoother overall win rates by civ, with more clustering around the midline. The lowest civ is also higher while the highest civ is lower.
Oh and also more content overall.

I think with this data the shoe is firmly on the other foot. Why Shouldn’t 4UC be integrated if it adds more stuff while also making the game more balanced and consistent?

I think with this the only real impediment is permissions and the work involved.

hokath · May 10, 2024

Aaand I think I get it. Is this what AI pick when they found/enhance ?

Must be. The alternative explanation are they come from Byzantium but the numbers are too high for that.

Edit: It's interesting Byzantium never won a cultural victory.

I think with this data the shoe is firmly on the other foot. Why Shouldn’t 4UC be integrated if it adds more stuff while also making the game more balanced and consistent?

Very much so. I don't feel 4UC gets anywhere near being out of scope, and the absence of any glaring balance issues confirms this. The way it slots onto the existing VP without taking away from it is the ideal form for a modmod that is to be integrated.
Compare with More Wonders--which I love and cherish--where a system gets totally overhauled. It's hard to see how this could ever win a supermajority when it completely changes key aspects of the wonders concept within the game.

azum4roll · May 11, 2024

pineappledan said:
Fewer Time victories, and more domination victories.
Slightly shorter games overall, but also eliminated sub-turn 300 wins.
Smoother overall win rates by civ, with more clustering around the midline. The lowest civ is also higher while the highest civ is lower.
Oh and also more content overall.

I think with this data the shoe is firmly on the other foot. Why Shouldn’t 4UC be integrated if it adds more stuff while also making the game more balanced and consistent?

I think with this the only real impediment is permissions and the work involved.

The DLL integration and AI, plus some of the more janky abilities (very subjective) are why I object to it.

Also way too many recon UUs!

pineappledan · May 11, 2024

azum4roll said:
Also way too many recon UUs!

With the polynesia change there will be 3 base VP recon UUs
4UC adds another 3: Greece, Inca, and Maya.
6 recon UUs out of 86. Calm down.

andersw · May 11, 2024

Is there any noticable difference in how fast victories come with 4UC, ie does the power it adds make the game go faster in number of turns?

pineappledan · May 11, 2024

There are victory timing box plots for with/without 4UC in the OP

Edit: check the victory type comparison tab

FinMc · May 11, 2024

azum4roll said:
The DLL integration and AI, plus some of the more janky abilities (very subjective) are why I object to it.

Also way too many recon UUs!

What do you mean by "DLL integration and AI"?

zeofig · May 13, 2024

Very cool data as usual.

I would really love to see data decoupling AI flavours from abilities. I think it would be really interesting. I suspect that some of the worst AIs are being crippled by their flavours which somehow contrive to make them play ineffectively, moreso than their civ ability set (Aztecs for example are extremely powerful in human hands, with or without 4UC). By flavours I mean personalities, policy tendencies, expansionism tendencies etc (anything apart from the actual civ bonuses). This kind of data would be a great way to balance part of the game through AI stats, because it takes out a bunch of the variables. Plus it's balancing something that's specific to the AI (flavours). It would make other AI-based balancing more objective since these flavours could be balanced first/independently.

Edit: to be more clear, I mean doing AI games with all civ abilities turned off. Or the opposite way: AI games where all AI players have identical flavours/personalities.

CppMaster · May 14, 2024

But AI personality is a part of AI. Without it, it would be neither AI or human.

pineappledan · May 14, 2024

Page 13 of Vern's report on the base game did indeed find that the AI war flavours and AI diplomacy flavours were strongly predictive of win rate.

For people unfamiliar with PCAs, 38% on a single principal component (the x axis) is enormous.

Spoiler :

I wrote a paper on a 650 AI Game data analysis (VP 4.5)

Some of you may be familiar with the AI autoplay game analyses I have been doing in this forum in the past. This semester, I expanded the scope of this initiative to a graduate data science course project where I explored additional topics and provided a written analysis that I hope you will...

forums.civfanatics.com

Recursive · May 14, 2024

zeofig said:
Very cool data as usual.

I would really love to see data decoupling AI flavours from abilities. I think it would be really interesting. I suspect that some of the worst AIs are being crippled by their flavours which somehow contrive to make them play ineffectively, moreso than their civ ability set (Aztecs for example are extremely powerful in human hands, with or without 4UC). By flavours I mean personalities, policy tendencies, expansionism tendencies etc (anything apart from the actual civ bonuses). This kind of data would be a great way to balance part of the game through AI stats, because it takes out a bunch of the variables. Plus it's balancing something that's specific to the AI (flavours). It would make other AI-based balancing more objective since these flavours could be balanced first/independently.

Edit: to be more clear, I mean doing AI games with all civ abilities turned off. Or the opposite way: AI games where all AI players have identical flavours/personalities.

Exercise caution with interpreting this data. The flavor system has bugs and issues with it. In particular the AI handling grand strategy / victory pursuit needs an overhaul, and the non-diplomacy flavors' effects are not well understood.

azum4roll · May 14, 2024

Policy picker (and likely other pickers) needs to be reworked too. It's full of arbitrary numbers for each ability.

pineappledan · May 15, 2024

Two notes about these games is that they contain at least 2 bugs that affect civ balance:
- Austria’s great person rate bonus per marriage was doubled to 30%
- Polynesia’s ability to build fishing boats with embarked units was unusable by the AI.

Both of these bugs have now been fixed for later versions, but Read the stats for both of those civs with that knowledge in mind.

Civ	VP Rank	4UC Rank	Rank Change	VP Winrate	4UC Winrate	Winrate Change
Siam	9	1	8￪	20	27.4	7.4⇈
Austria	1	2	-1￬	31.9	26.2	-5.6⇊
Germany	4	3	1￪	24.4	24.5	-
Korea	8	4	4￪	20.4	23	2.6￪
Poland	13	5	8￪	18.2	22.2	4.0￪
Arabia	5	6	-1￬	21.5	21.1	-
Songhai	12	7	5￪	18.8	20.5	1.7￪
China	3	8	-5￬	24.6	20.4	-4.2￬
India	18	9	9￪	13	20.2	7.1⇈
The Netherlands	2	10	-8￬	27.5	20	-7.5⇊
Byzantium	22	11	11⇈	10.9	17.6	6.7⇈
The Maya	10	12	-2￬	19.3	17.2	-2.1￬
The Inca	14	13	1￪	17.9	16.8	-1.1￬
Greece	15	14	1￪	16	16	-
Carthage	37	15	22⇈	5.2	15	9.8⇈
Babylon	24	16	8￪	10.5	14.9	4.3￪
Egypt	23	17	6￪	10.7	14.7	4.0￪
Venice	7	18	-11⇊	20.7	14.5	-6.2⇊
Polynesia	29	19	10⇈	7.8	13.3	5.5⇈
Morocco	11	20	-9￬	18.9	13.2	-5.7⇊
Portugal	17	21	-4￬	13.5	12.2	-1.3￬
Russia	26	22	4￪	9.9	11.8	1.9￪
The Ottomans	6	23	-17⇊	20.8	11.5	-9.3⇊
The Zulus	32	24	8￪	6.9	11.4	4.5￪
The Celts	19	25	-6￬	12.5	10.1	-2.4￬
Brazil	20	26	-6￬	11.3	9.7	-1.6￬
Sweden	25	27	-2￬	10.3	9.5	-
Japan	38	28	10⇈	4.7	8.8	4.1￪
Assyria	21	29	-8￬	11.2	7.2	-4.0￬
Indonesia	16	30	-14⇊	14	7	-7.0⇊
England	34	31	3￪	6.7	7	-
Ethiopia	31	32	-1￬	7.1	6.7	-
Denmark	35	33	2￪	6.2	5.7	-
America	36	34	2￪	5.7	5.5	-
France	41	35	6￪	1.6	5.4	3.9￪
Rome	33	36	-3￬	6.7	5.1	-1.6￬
Persia	27	37	-10⇊	8.1	5.1	-3.1￬
The Iroquois	30	38	-8￬	7.8	4.9	-2.9￬
The Shoshone	39	39	-	3.4	4.4	-
Mongolia	42	40	2￪	1.5	3.8	2.3￪
The Huns	43	41	2￪	0.7	3.6	2.9￪
The Aztecs	28	42	-14⇊	7.9	2	-6.0⇊
Spain	40	43	-3￬	1.7	0.8	-

650 AI Game 4UC Stats / Analysis

Warlord

Warlord

Warlord

Prince

Prince

Deity

King

Deity

Emperor

Lost the game

Deity

Emperor

Deity

Warlord

Warlord

Deity

Deity

Already Looping

Lost the game

Deity

Similar threads