Sullla's AI Survivor Season Seven - Championship Game Thread

[Saxo Grammaticus]

Or we can run duplicate tests of clean and flawed versions just in case...

I appreciate having an abstract championship with a more even playing field as it gives a greater sense of the individual competence/character of the AI. That said, I find this map can often boil down to intense religious conflict, especially with a peace weight spread. Though any of these leaders can found an early religion, Darius, Louis, Ramesses, and Saladin would seem more likely to emerge religious leaders. None of these AI are obvious pushovers, so I would imagine diplo heavily factoring into FTD.

Rather than rolling a random map and balancing resources, perhaps a mapmaker could copy and paste together a new abstract map each season. One could even draw inspiration from the six finalists, e.g. inland starts for leaders without fishing, though that could delay the championship game.

 

[Keler]

Last game Darius was the clear winner after ~100t already.
I think this map at least avoids those scenarios..

Exact reason why I like this championship map over heavily unfair pangea maps. Otherwise game excitement ends so fast.
Balanced is probably a better map script among other default maps.

 

[civac]

Don't know about the current state of the map. But in previous most errors were in a location with seafood directly adjacent to capitals. Some had 2, some had 1 seafood resource and the location of the resources may have differed as well.

 

[The Rook]

Rather than rolling a random map and balancing resources, perhaps a mapmaker could copy and paste together a new abstract map each season.

Um... yeah I could make Sullla a map, and I'll run the AI autoplay at least 20 times just to make sure it's, you know, super well balanced and stuff.

 

[danjuno]

Just copypasting my form response

A very odd crowd for this years championship game, far too unpredictable. I'm gonna say that Mehmed shatters his lance on Darius with a pre-construction rush, Darius eats him and snowballs, while Churchill backdoors another Silver.

 

[Plains-Cow]

Um... yeah I could make Sullla a map, and I'll run the AI autoplay at least 20 times just to make sure it's, you know, super well balanced and stuff.

Make every resource tile a plains-cow tile.

 

[danjuno]

Make every tile period a plains cow tile!

 

[Kaitzilla]

I found the old Season 6 championship map.
If it is the same map as Season 7 championship, should be able to look for any map errors.

Civ4 AI Survivor 6

Spoiler :

Have to replace a Civ 4 game file to open it.

Note that these files were generated with the Apostolic Palace removed from the gameplay and therefore will not load normally onto a standard copy of Civ4.
You will need to make a change to one of the XML files in order to run these saves.
I've included the edited version of this file "CIV4BuildingInfos.xml" which can be found on your local hard drive under Program Files\Firaxis Games\Sid Meier's Civilization 4\Beyond the Sword\Assets\XML\Buildings.

Make sure to save the original and put it back before playing normal Civ 4 again.

 

[NickGabben]

How would you improve that aluminium anyway, with a workboat or with aquaman workers?

 

[0urBall]

Am late to this particular party so haven't seen all of the other seasons. But if Mehmed doesn't win, does that mean each season has had a different winner? And, if that's the case, has anyone suggested some sort of "Champion of Champions" match as there would be 7 Civs?

 

[Keler]

I think you guys shouldn't worry about errors,Sulla fix them all the time before playing live. Consider it fair and same.

the map is publicly available, so there's nothing to stop anyone from carrying out their own tests and knowing the best percentages in advance

yeah about that last season Mansa Musa won so our tests won't really guarantee help us earn more points.


For example I autoplayed 10 times and had these results:

Winner Ramesses	Runner up Mehmed	FTD Darius	Turn 299	Space
Winner Darius	Runner up Saladin	FTD Louis	Turn 314	Space
Winner Darius	Runner up Churchill	FTD Louis	Turn 339	Space
Winner Darius	Runner up Ramesses	FTD Mehmed	Turn 329	Space
Winner Ramesses	Runner up Louis	FTD Mehmed	Turn 273	Culture
Winner Ramesses	Runner up Darius	FTD Louis	Turn 324	Space
Winner Churchill	Runner up Mehmed	FTD Darius	Turn 383	Space
Winner Ramesses	Runner up Darius	FTD Mehmed	Turn 320	Culture
Winner Ramesses	Runner up Saladin	FTD Mehmed	Turn 327	Culture
Winner Darius	Runner up Ramesses	FTD Mehmed	Turn 306	Space

Maybe if I autoplayed 10 more times I would witness someone else's victory.

Rather than rolling a random map and balancing resources, perhaps a mapmaker could copy and paste together a new abstract map each season

Agreed, by the end of twitch video Sulla mentions about creating new map for future seasons. Hopefully we see that in a year or two.

 

[Fippy]

But aren't those autoplays a bit like watching movies with a spoiler?

 

[Keler]

But aren't those autoplays a bit like watching movies with a spoiler?

True
How about we rank default and custom map scripts instead of AIs,leader traits and civilization starting techs.
Which random non geometric map scripts are fairer?

 

[Thrasybulos]

Was away on vacation for a while, so missed some of the fun there.

AI Survivor has to deal with contradictory constraints.
It's an AI competition and a ranking attempt : you want "fair" and balanced maps.
It's a prediction contest : you want unbalanced maps (ideally not in an obvious fashion, though) so that map reading skills apply.
It's a live show : you want drama brought by completely random and unpredictable elements (barbs, UN, ...).

Seems to me the first two are totally irreconcilable for the Championship game, unfortunately.

 

[The Rook]

yeah about that last season Mansa Musa won so our tests won't really guarantee help us earn more points.

Ironically, I think your example shows why balanced maps are a bad option in the context of a picking contest. Let's see what the alternate histories say:

It turns out that the majority were correct with their picks of HC to win (8 times!) and Mansa as FTD (13 times!!!). This was an easy assessment, as Mansa had the highest peace weight in a low peace weight field, and HC clearly had the best Survivor history. Map considerations? All AIs had the same start and could potentially share a border .

In contrast, Darius won both of his starts this year very convincingly against low peace weight fields. This would be unlikely on a balanced map, but the merits of his position were enough to overcome this. While some may say this was unfair, I still wrongly picked Willem to win the first game, and I see that you incorrectly picked Qin to win the playoff, so it clearly wasn't obviously unfair . Also, the alternate histories show that Hatty had an unfairly good start, but that didn't stop most people (including me) from backing Cathy to win!

Just to clarify, I don't equate imbalance with unfairness, sometimes imbalance can improve fairness by giving a terrible AI a chance to do something. Most importantly we want maps where our "reading skills" apply (As Thrasybulos mentioned). It seems to me that there is an inverse relationship between reading skills and balance for picking purposes.

For example I autoplayed 10 times and had these results:

Winner Ramesses	Runner up Mehmed	FTD Darius	Turn 299	Space
Winner Darius	Runner up Saladin	FTD Louis	Turn 314	Space
Winner Darius	Runner up Churchill	FTD Louis	Turn 339	Space
Winner Darius	Runner up Ramesses	FTD Mehmed	Turn 329	Space
Winner Ramesses	Runner up Louis	FTD Mehmed	Turn 273	Culture
Winner Ramesses	Runner up Darius	FTD Louis	Turn 324	Space
Winner Churchill	Runner up Mehmed	FTD Darius	Turn 383	Space
Winner Ramesses	Runner up Darius	FTD Mehmed	Turn 320	Culture
Winner Ramesses	Runner up Saladin	FTD Mehmed	Turn 327	Culture
Winner Darius	Runner up Ramesses	FTD Mehmed	Turn 306	Space

Maybe if I autoplayed 10 more times I would witness someone else's victory.

I don't think 10 tests are really enough. Anyway, I'm sure I don't need to argue the value of testing with the person who published the extensive: AI Autoplay Ranking- testing with fair maps

I agree with Fippy that running tests is like seeing spoilers, not only in terms of results, but also you will see certain patterns continually repeated. If someone painstakingly recreated Sullla's random maps and won the overall picking contest through their consistency, then I'd at least argue they had (more than) earned it. However, it doesn't feel quite in the spirit of a picking contest the idea of downloading a map and running autoplay to determine the best picks. BTW, this is not an attack on you, anyone can test the finalists on the Commodore map (and I expect several pickers are!).

So while I might hold a minority opinion that we should have a random map in the final (with potential re-rolls if it is obviously unfair), I will rigorously defend it.

 

[Fippy]

Well i'm surprised by Keler's results..Darius borders backstabber Lou & warmonger Mehmed and still has so much success.

 

[Plains-Cow]

Well i'm surprised by Keler's results..Darius borders backstabber Lou & warmonger Mehmed and still has so much success.

Could it be the Year of Persia?

 

[lymond]

Could it be the Year of Persia?

Hmmmm...I thought it was the Year of the Cow

 

[Plains-Cow]

Hmmmm...I thought it was the Year of the Cow

Not until @Sullla puts a plains-cow tile in the championship game.

...also, every year is the Year of the Cow!

 

[The Rook]

I've looked at the map and have noted as many errors I could find (some have been mentioned already (AcaMetis & Fippy)). Unfortunately, I won't be able to watch the game live, but hopefully someone here could notify Sullla about the necessary corrections when the stream starts. Churchill seems to have the best deal, minus several fish resources, but has the bonus of coastal aluminium!

1. Coastal Aluminium
2. Missing fish
3. Missing fish

4. Missing fish

5. Replace wine with cow
6. Louis "might" be missing a scout.

Concerning future finals - I've been trying to think of ways to make balanced maps more interesting from a map reading perspective. How about introducing an unusual theme each season. For example, you could have the plains cow final, where every non-resource plains tile on the map is a plains cow. Would this cause a balanced map to play differently? Perhaps the huge production bonus would help the warmongers, or would it just cause them to field massive armies of terminally obsolete units? Would the plains cows slow the tech pace down with low commerce, or will the food and production bonuses mitigate this through faster wonder spamming and research building? Just some thoughts.