Civ4 AI Survivor Season Seven: Back AI-gain



Introduction

After a hiatus of more than a year (coinciding with the birth of our first child), the calendar has turned to 2023 and it's time to launch another season of Civ4 AI Survivor! This will be the seenth time in all that we're embarked on this venture. Back in Season One, I was still figuring out the format and running everything via forums posts at Realms Beyond. It was a bit of a rough process even if it was still fun. For Season Two, I switched over to Livestreaming the matches in real time, and that made things infinintely more fun for everyone with the viewers getting to comment on the action in real time. It's not an exaggeration to say that the running commentary on the Livestream was often much more entertaining than whatever the AI leaders happened to be doing at the moment. Season Three saw more refinements to the process, with maps selected ahead of time and blindly assigned to the leaders for purposes of fairness, and online straw polls/predictions recorded before the matches and included in a series of written previews. Season Four initiated the adoption of the correct starting techs for each civilization along with the implementation of a second AI-controlled observer civ so that we could watch the voting in the Apostolic Palace without fear of a religious victory taking place. Season Five ditched the Apostolic Palace and saw the largest turnout yet in terms of the community. Aided no doubt by the COVID-19 pandemic that left many people working from home, Season Five of AI Survivor exploded in popularity to reach an average of 250 contest submissions and hundreds of viewers tuning into the Livestream. At its height, I was averaging 150+ viewers across the length of a seven hour stream and topping 250 viewers at its peak. Season Six added a fantasy sports contest with auction bidding of the AI leaders running in parallel with the normal scoring competition for each individual game. Let's hope that we can keep things going strong with an entirely new competition here for Season Seven.



The basic tournament structure will remain the same from past seasons. The 52 AI leaders in Civilization 4 get randomly sorted into eight opening round games, half of them with 6 leaders and half of them with 7 leaders. Two leaders will advance to the playoff round from each match, the winning leader and the non-winner with the highest in-game score. AIs who survive the match but do not advance will get one final opportunity to make the playoffs in a Wildcard game played after the opening round matches. The Wildcard game has had as many as a dozen leaders and as few as six leaders in past years and it can get a little bit weird as a result. (The Wildcard game is also the only one that we run with Raging Barbarians turned on, and yes, we have seen an AI leader lose their capital city to the barbs!) This creates our field of 18 leaders sorted into three playoff games of 6 AIs apiece. Finally, we feed the top two leaders from each playoff game, the winner and the highest scoring non-winner, into a championship game and crown an overall winner of the competition.

For Season Four, we tried out a seeding system akin to what the World Cup uses when drawing teams into its groups. This proved to be a big success and we'll be continuing it for Season Seven. The top eight AI leaders based on past performance will be placed in Pool 1, and the next eight AI leaders based on past performance will be placed in Pool 2. Each of the opening round games will have one leader drawn from Pool 1 and one leader drawn from Pool 2. The remaining 36 AI leaders in Civ4 will be unseeded and can be randomly drawn into any opening round game. Here's a list of which leaders fall into which category:




AI leader rankings are based on our customized scoring system of past performances; click here for a link to the full ranking as of the end of Season Six. AI leaders score 5 points for a first place finish, 2 points for a second place finish, and 1 point for each elimination of another leader. While this is a purely arbitrary system, it seems to provide an effective metric for evaluating prior seasons of the competition. The idea behind the seeding system is to stop the top AI leaders from hitting one another in an opening round "group of death" match while still preserving the random draw element that makes this event so much fun. Choosing to pick pools of leaders ranked 1-8 and 9-16 is of course another arbitrary decision but there's some logic to having one top leader and another secondary leader in each of the eight opening round games. Going into Season Seven, there were no ties for any of the leaders right on the cusp of being in Pool 1 or Pool 2, making it easy to sort everyone into their respective categories. Nice and neat this time around.

The game settings will continue over from the last Civ4 AI Survivor competition. We use Deity AIs to speed up the overall pace of the game and reach victory results sooner. The map script is the default Pangaea, designed to keep all of the AI leaders in close contact with one another. Prior to the Season Three competition, I went ahead and tested out some games with Continents or Archipelago setups, and they just didn't work that well. The Civ4 AI doesn't really know how to run naval invasions, and there were lots of wars that just did nothing because the AIs couldn't reach one another. They need to be on the same continent from what I've found, and Pangaea works well enough for that. We use Normal game speed and Standard sized maps, with the one exception of dropping the sea level to "Low" setting for the games with 7 AIs to create a little more room for the extra competitior.

For the game options, the biggest effect on the gameplay is turning off Tech Trading. I'll repeat here what I wrote in the setup phase from past years: "When tech trading is left on, the AIs tend to stay bunched together with no one ever getting too far ahead or falling too far behind. With everyone fielding the same units, it becomes difficult for the AIs to attack one another successfully, and generally makes for less interesting games. But when tech trading gets switched off, all of the AIs have to sink or swim based on their own research efforts. This spreads them out much more widely on the tech tree, and once the laggards fall a full generation behind in military tech, they tend to get swallowed up by their more advanced rivals." With tech trading removed, we frequently see some AIs fall an entire era or more behind in technology, which prevents would-be conquerers from ignoring their economies and trading their way to tech parity. A balanced approach tends to work the best in these games, with alternating periods of warring to expand followed by peaceful consolidation. Woe to those who fall behind in military tech in this competition - it nearly always ends badly. We've also increasingly noticed the different research preferences of the various AI leaders in recent years, with the warmongers overly obsessing on military techs and some of the peacenik AIs refusing to pick up Rifling even as they're dying to foreign invaders. Turning on tech trading would remove this important aspect of the competition and make the overall setup a lot less interesting.

Otherwise, we turn off Vassal States to prevent the AIs from surrendering to one another. If they can vassal when losing wars, hardly any of the AIs ever get eliminated, and that's boring. We want to see bloodthirsty games! No escape, no mercy. We also turn off Events because they're kind of a Single Player thing, and the AI doesn't understand what they do. Similarly, we turn off the goody huts to prevent unwanted prizes from getting handed out to the AIs. We've been moving towards a less random and more standardized setup for the past few years and this seemed like an obvious next step. No more cases of someone getting lucky and popping a critical tech out of a hut on the first turn of the game - now they need to research their own stuff like everyone else. We also use Aggressive AI to spice things up and increase the number of wars. The big thing that Aggressive AI does is remove a flag that will stop the AIs from declaring certain wars. Normally they will not attack someone who has more than double their military power. Turning on Aggressive AI removes that check in their coding logic. This leads to occasionally suicidal war declarations based on diplomatic relations, and that's exactly the kind of thing we want to see!

The biggest change for Season Five was the removal of the Apostolic Palace (quite literally; I deleted the wonder from the XML files on my local computer). The Apostolic Palace had been the most controversial aspect of prior seasons from a gameplay perspective, with the AI civ that controls the Apostolic Palace often using it to shut down literally every invasion launched against them, or forcing everyone in the game to declare war on some hapless target. The argument in favor of the AP was that it introduces a wild and unpredictable element of chaos into each game which could be a lot of fun to watch. The argument against the AP was that a single wonder shouldn't have such a massively disproportionate effect on the rest of the gameplay, and in particular, it shouldn't become a magical shield that could shut down all aggression for hundreds of turns on end. I decided to run a poll question asking the community whether they wanted to see the inclusion of the Apostolic Palace for Season Five and the community voted decisively to remove it by a greater than 2:1 margin. The general consensus following the end of the season was that the gameplay had improved noticeably by leaving it out of the competition. We will not be bringing it back anytime soon.

New Changes for Civ4 AI Survivor Season Seven

Since we've run six full seasons in the past, we already have a pretty good idea of what works and what doesn't work for this format, and further changes to the AI Survivor setup should be minimal moving forward. We're changing the climate setting back to the default "Temperate" option after experimenting with some of the other climates in recent years. This should create more balanced maps with fewer huge patches of desert or ice or jungle. We'll also continue to use the "Choose Religions" options so that we can avoid getting Buddhism and Hinduism as the dominant religions in every game.

I've also learned over the years about the best ways to run the AI Survivor event from an organizational standpoint. Following the pattern that began in Season Six, I'm planning to hold the draw for all of the opening round games at the same time, on Livestream several weeks before the first match begins (currently planned for 17 March 2023 at the usual noon Livestream time). I'll pick out all of the maps for the games ahead of time and then we'll fill in the leaders from the initial draw. This is largely being done to help save time during the season; by drawing the leaders and prepping the maps before the season starts, I won't have quite as much work to do on a week-by-week basis. I'll also be able to post the previews for all eight opening round games immediately following the draw to again save time and let the community get an early look at the upcoming matchups. Aside from helping to cut down on some of the behind-the-scenes work that goes into AI Survivor, doing the early draw also allows us to implement one of the more popular community features:



Fantasy Civ4 AI Survivor

A number of people previously asked about instituting some kind of fantasy feature in past seasons and we were finally able to test it out for the first time in Season Six. We'll follow the same setup as before: the leader draw will take place several weeks before the first match of the season. I will then post all of the opening round previews so that everyone has time to study the upcoming games. One week before the start of the season, I plan to hold an auction draft on Livestream with a group of other community members as we bid on which leaders we want to fill out our fantasy AI Survivor teams. I think that this would work with as few as four or five other individuals but my preference would be to get eight volunteers for a league of eight fantasy teams. I entered the fantasy contest (and won it!) in Season Six but my preference would be to find eight participants for Season Seven so that I can remain in an impartial moderator role when conducting the auction draft. We'll run through the AI leaders and see who can put the best team together for the rest of the season.

If you're interested in taking part in the auction draft, here's a bit more information:

* Scoring will be based on our standard AI Survivor points system: 5 points for a first place finish, 2 points for a second place finish, and 1 point for each elimination of another leader. The further that the leaders on your team make it, the more chances they'll have to score points.
* Unlike the normal AI Survivor points system, players will also score 2 points for having a "First to Die" leader on their team. This should make the bidding much more interesting!
* There are no "positions" to worry about or a limit to the number of AI leaders that can fit on a team. Whether to have a small handful of favorites or a large group of underdogs is a choice that fantasy participants will have to make.
* To take part in the auction draft, participants will need to be available for 2-3 hours starting around noon (EST) on Friday, 31 March 2023. You will also need to join the Sullla Discord channel and a microphone is strongly preferred for voice communication.
* This is open to anyone interested who can be there for the live draft. I will feel free to give preference to longtime AI Survivor community members though; for example, noted superfan LinkMarioSamus has a spot if he wants it.
* Current signups: nabaxo, Myth, Slashin, cutHrAXys, Amicalola, jmie6, sunyujie, Henrik. Alternates: Kjotleik, Burninator1729

To get a better feel for what this looks like, here's the video recording for the auction draft that took place before the beginning of Season Six. I also put together a lengthy written analysis looking back at the fantasy results for Season Six after the full competition was over. I'll keep track of the fantasy teams and we'll see who winds up having the best season once the games start! We will also still be running our normal picking contest each week; see the details below for those who are unfamiliar with the process. The fantasy draft is more intended for the serious fans of AI Survivor (which we absolutely do have!) as opposed to the picking contest which requires minimal time investment.

Watching and Following Season Seven

Games will be Livestreamed on my Twitch channel linked below, with most games starting at noon local time (east coast of North America) on Fridays. I will archive the recordings on Twitch and post them on YouTube for those who have trouble accessing the Twitch videos. I'm also soliciting help once again in putting together the written reports on the games that took place, in the same fashion that we've done for the games from past seasons. Several folks pitched and submitted written reports for Season Six and it was a blast getting to see different spins on how the games played out. It will be a huge help if the community can continue to help shoulder the burden of doing the written reports and allow me to concentrate on streaming and administering the games. This is even more true now that we have our little baby at home requiring lots of love and care - anyone interested in volunteering to help out again?

For those who are new to our event, we run a picking contest for each game that serves as the biggest driver of interest in AI Survivor. We ask contest entrants to pick:

* The winning AI leader
* The second place AI leader
* The "First to Die" AI leader
* The victory type of the winner
* The victory date of the winner
* The total number of wars in the game

The closer that the pick gets to the correct answers, the more points that it scores. We track scores for both individual games and the overall competition so there's always a chance to win an individual round even for those who missed earlier matches. The picking contest system will continue to use a Google Forms submission, which was a massive timesaver in recent seasons as compared to originally compiling forum posts. The current bracket will always be visible online, and I'll do my best to update it here on my website, on my Twitch page, on the Sullla Discord channel, and at Realms Beyond. We'll have a forum discussion thread there, of course, which I'll link on this page once the competition has started. Most of the discussion has migrated over time to the Discord though which has its own dedicated channel for AI Survivor (and which remained highly active even during the long drought between seasons). Scoring remains the same from past years, with points for first place, second place, first to die, victory date/type, and total number of wars. With points for the top two AI spots along with victory date and number of wars, we had scoring drama right up to the end of most every game, even when the winner had long since been determined. Check out the links below for more information on the scoring contest and how to submit your entries.

With all that said, here's a big list of the relevant links that follows below. Enjoy the new season of AI Survivor!

Main Season Seven Webpage

Sullla Twitch Page

Sullla Discord Channel

Civ4 AI Survivor Season Seven: Leader Previews

AI Reference Guide from CivFanatics

Complete AI Survivor Archives

Next Game: Opening Round Game One

Schedule: Season Seven Leader Draw, 17 March 2023 at Noon EST
Schedule: Season Seven Fantasy Auction Draft, 31 March 2023 at Noon EST
Schedule: Season Seven Game One, Starts 07 April 2023 at Noon EST

This should be fun. 

You know, I realize what's a big flaw about the Warlord retort. It mainly gives two benefits: first, all units start with one level higher than the base, and second, they have a higher max level than normal. Now, the latter benefit is far stronger than the former, but can only really be exploited by Life via the Heroism spell. When you do that, the retort is well worth its cost in picks. But if you don't, all you get is a mild early game benefit on your units that fades away quite quickly. This explains why it feels so underwhelming when picking it for non Life games.

What do you think? How should this be fixed? Or is it fine as it is?

I'm curious about this, since I find this combination rather hard to get off the ground. Most of the early game spells don't synergize, outside of like Hell Hounds with repeated Phantom Warrior/Confusion casts, nor do they seem to excel at any aspect early on. Flame Blade and Resist Magic are excellent buffs, and Wall of Fire and Philosopher's Stone both make for amazing defenses, but conquering things seems tricky, as neither school provides particularly strong defenses for units or offensive spells against armies until the Rare tier. I'm particularly interested in hearing about successful games with this combination in the Warlords mod, but games on vanilla CoM 2 are fine too.

Hi,

I would like to suggest and get suggestions about a quick combat results calculation for the AI and perhaps also for the game quick combat suggestion/execution for the player.
So here is my partial suggestion:

let's take a simple example:
two sides A,B no special or range abilities no spell points only 1 figure

let:
dmgPot = damage potential
defPot = defense potential
HP = hit points

AdmgPot = Aattack*Atohit
BdmgPot = Battack*Btohit
AdefPot = Adefense*Atodef,
BdefPot = Bdefense*Btodef,

ARoundDmg = (BdmgPot-AdefPot) - how much damage A sustained every round
BRoundDmg = (AdmgPot-BdefPot) - how much damage B sustained every round

in case BRoundDmg == 0 BRoundDmg = some very small number 10^-10? to prevent division by zero

numRounds = Minimum(25, BHP/BRoundDmg , AHP/ARoundDmg)))

BattleADmg = numRounds*ARoundDmg - how much damage A got by the end of the battle
BattleBDmg = numRounds*BRoundDmg - how much damage B got by the end of the battle

in case we have two stacks of units AStack, BStack:
no special or range abilities no spell points, 1 figure and no one tries to run away on the battle map,
side A will pick from team B the unit to attack with the greatest (BdmgPot/BdefPot)/BHP lets call this unit C
and will attack it with his unit with the greatest
AdamagePrecentDiff =
(AdmgPot-CdefPot)/CHP - (CdmgPot-AdefPot)/AHP

when (AdmgPot-CdefPot)/CHP is how much damage A did to C
and (CdmgPot-AdefPot)/AHP is how much damage A received

when AdamagePrecentDiff is negative or the total units AdamagePrecentDiff is negative =
we kill less percentage of the enemy than he kills us
we might consider running away or avoiding combat on the battlefield.

This can be expanded to include figures, spells speed, resistance etc..

Who will be the new king of diplo? Stay tuned!

You bet I'm reusing songs from PB67. I had been saving some of them for a long time. Plus this song is GREAT!




PB67 map problems were a good example of diplomacy being important. SD and I started at dagger distance and yet we didn't war. I'll be interested to see how this develops.

I'm pulling hard for block pick method. For same reason as PB65. I don't people to get whatever busted combo they have thought up with America. I was correct that Cornflakes had a better idea probably than I did, but I like simple so eh.

If I don't get America we will see. Tarkeel said 145 / player but feels smaller, so I'm going to play picks like its 120/player. Expect me to mostly do picks based on past preferences. There is a chance depending on start I might go financial instead of org. A more balanced map and diplo MAY* mean it goes long. I'm once again torn on Ind in a diplo game. Two people chose last diplo game. If we have at least 3 neighbors though I think any early wonder may just be a trap to invite people to nibble bits off your land. But with only 5 being the only Ind leader along with a little room could be very good. Something like Ind/pro Mali has been on my list for a while. Cre/Fin Mali has also been on my list. Basically I really want to play Mali again, but will be start and feel dependent although if it is snake and I can I will pick America if I have a seafood start (or low tech requirements).

text goes here

Please Tarkeel give us all the corn and gold and stone, and may our neighbors have no horse, amen 

Well, well, well.

And there I thought PB67 is just dead and the follow up game has been ghosted. I have to say the end of PB67 did not help raise my enthusiasm for playing in an MP again, but I promised to play in this game and will try to make the best of it. After all this is still a diplo game and might turn out interesting. What are you saying the player, with whom you can have the best chat has left and turned mapmaker. Well shit, this might turn out to be the most boring diplo game looking at rest of mutes in this. Maybe I'm lucky and Mjmd is my neighbor. At least he is talkative.

Another reason why I'm still very annoyed about the ending of PB67 is that I had a lot of fun with posting those music clips. I still don't know how to continue with that after I used up a good chunk of my collection of songs. Well if I'm not mistaken we will use the same snake pick/civ/leader combo as in PB67 and the same starting screenshot, so I may just not have to report about all the early stuff as much. For those that do not know what I'm talking about with music here is what I wrote in PB67:

The reporting in PB59 was very enjoyable for me so I will keep it that way. Meaning:

• I won't report every single turn
  
• Same crazy micro managment spreadsheet
  
• (Almost) no sims: I will actually run almost no sims this game, because my spreadsheet can replace them fully. I remember back in PB52 I stopped running sims about T50 or so, but I used my spreadsheet up to T110 or so. Expect more of that.
  
• C&D: This is very tedious for me and the benefit is smaller. Even if we play with score I won't bother with this analysis.
  
• Of course this being a diplo thread I will attempt to write down the important diplo stuff down as best as possible.